id sid tid token lemma pos fcis-4086 1 1 frontiers frontier NOUN fcis-4086 1 2 in in ADP fcis-4086 1 3 computing computing NOUN fcis-4086 1 4 and and CCONJ fcis-4086 1 5 intelligent intelligent ADJ fcis-4086 1 6 systems system NOUN fcis-4086 1 7 issn issn VERB fcis-4086 1 8 : : PUNCT fcis-4086 1 9 2832 2832 NUM fcis-4086 1 10 - - SYM fcis-4086 1 11 6024 6024 NUM fcis-4086 1 12 | | NOUN fcis-4086 1 13 vol vol NOUN fcis-4086 1 14 . . PROPN fcis-4086 2 1 2 2 NUM fcis-4086 2 2 , , PUNCT fcis-4086 2 3 no no INTJ fcis-4086 2 4 . . NOUN fcis-4086 2 5 2 2 NUM fcis-4086 2 6 , , PUNCT fcis-4086 2 7 2022 2022 NUM fcis-4086 2 8 58 58 NUM fcis-4086 2 9 research research NOUN fcis-4086 2 10 on on ADP fcis-4086 2 11 news news NOUN fcis-4086 2 12 text text NOUN fcis-4086 2 13 classification classification NOUN fcis-4086 2 14 based base VERB fcis-4086 2 15 on on ADP fcis-4086 2 16 rcnn rcnn PROPN fcis-4086 2 17 xudong xudong PROPN fcis-4086 2 18 wang wang PROPN fcis-4086 2 19 asterfusion asterfusion PROPN fcis-4086 2 20 data data PROPN fcis-4086 2 21 technologies technologies PROPN fcis-4086 2 22 co. co. PROPN fcis-4086 2 23 , , PUNCT fcis-4086 2 24 ltd ltd PROPN fcis-4086 2 25 . . PROPN fcis-4086 2 26 , , PUNCT fcis-4086 2 27 china china PROPN fcis-4086 2 28 abstract abstract NOUN fcis-4086 2 29 : : PUNCT fcis-4086 2 30 one one NUM fcis-4086 2 31 of of ADP fcis-4086 2 32 the the DET fcis-4086 2 33 popular popular ADJ fcis-4086 2 34 directions direction NOUN fcis-4086 2 35 of of ADP fcis-4086 2 36 natural natural ADJ fcis-4086 2 37 language language NOUN fcis-4086 2 38 processing processing NOUN fcis-4086 2 39 is be AUX fcis-4086 2 40 text text NOUN fcis-4086 2 41 processing processing NOUN fcis-4086 2 42 , , PUNCT fcis-4086 2 43 which which PRON fcis-4086 2 44 is be AUX fcis-4086 2 45 an an DET fcis-4086 2 46 effective effective ADJ fcis-4086 2 47 method method NOUN fcis-4086 2 48 used use VERB fcis-4086 2 49 to to PART fcis-4086 2 50 manage manage VERB fcis-4086 2 51 data datum NOUN fcis-4086 2 52 , , PUNCT fcis-4086 2 53 and and CCONJ fcis-4086 2 54 it it PRON fcis-4086 2 55 has have AUX fcis-4086 2 56 developed develop VERB fcis-4086 2 57 rapidly rapidly ADV fcis-4086 2 58 after after ADP fcis-4086 2 59 the the DET fcis-4086 2 60 rise rise NOUN fcis-4086 2 61 of of ADP fcis-4086 2 62 artificial artificial ADJ fcis-4086 2 63 intelligence intelligence NOUN fcis-4086 2 64 . . PUNCT fcis-4086 3 1 however however ADV fcis-4086 3 2 , , PUNCT fcis-4086 3 3 text text NOUN fcis-4086 3 4 classification classification NOUN fcis-4086 3 5 based base VERB fcis-4086 3 6 on on ADP fcis-4086 3 7 traditional traditional ADJ fcis-4086 3 8 machine machine NOUN fcis-4086 3 9 learning learn VERB fcis-4086 3 10 algorithms algorithm NOUN fcis-4086 3 11 encountered encounter VERB fcis-4086 3 12 bottlenecks bottleneck NOUN fcis-4086 3 13 in in ADP fcis-4086 3 14 the the DET fcis-4086 3 15 development development NOUN fcis-4086 3 16 process process NOUN fcis-4086 3 17 , , PUNCT fcis-4086 3 18 and and CCONJ fcis-4086 3 19 after after ADP fcis-4086 3 20 deep deep ADJ fcis-4086 3 21 learning learning NOUN fcis-4086 3 22 , , PUNCT fcis-4086 3 23 a a DET fcis-4086 3 24 multilevel multilevel ADJ fcis-4086 3 25 neural neural ADJ fcis-4086 3 26 network network NOUN fcis-4086 3 27 , , PUNCT fcis-4086 3 28 was be AUX fcis-4086 3 29 introduced introduce VERB fcis-4086 3 30 , , PUNCT fcis-4086 3 31 it it PRON fcis-4086 3 32 was be AUX fcis-4086 3 33 soon soon ADV fcis-4086 3 34 widely widely ADV fcis-4086 3 35 used use VERB fcis-4086 3 36 in in ADP fcis-4086 3 37 natural natural ADJ fcis-4086 3 38 language language NOUN fcis-4086 3 39 processing processing NOUN fcis-4086 3 40 problems problem NOUN fcis-4086 3 41 , , PUNCT fcis-4086 3 42 making make VERB fcis-4086 3 43 many many ADJ fcis-4086 3 44 branches branch NOUN fcis-4086 3 45 of of ADP fcis-4086 3 46 the the DET fcis-4086 3 47 field field NOUN fcis-4086 3 48 develop develop VERB fcis-4086 3 49 further far ADV fcis-4086 3 50 , , PUNCT fcis-4086 3 51 and and CCONJ fcis-4086 3 52 now now ADV fcis-4086 3 53 deep deep ADJ fcis-4086 3 54 learning learning NOUN fcis-4086 3 55 is be AUX fcis-4086 3 56 the the DET fcis-4086 3 57 mainstream mainstream NOUN fcis-4086 3 58 model model NOUN fcis-4086 3 59 in in ADP fcis-4086 3 60 the the DET fcis-4086 3 61 field field NOUN fcis-4086 3 62 of of ADP fcis-4086 3 63 text text NOUN fcis-4086 3 64 classification classification NOUN fcis-4086 3 65 . . PUNCT fcis-4086 4 1 in in ADP fcis-4086 4 2 this this DET fcis-4086 4 3 paper paper NOUN fcis-4086 4 4 , , PUNCT fcis-4086 4 5 we we PRON fcis-4086 4 6 first first ADV fcis-4086 4 7 introduce introduce VERB fcis-4086 4 8 the the DET fcis-4086 4 9 general general ADJ fcis-4086 4 10 process process NOUN fcis-4086 4 11 of of ADP fcis-4086 4 12 text text NOUN fcis-4086 4 13 classification classification NOUN fcis-4086 4 14 , , PUNCT fcis-4086 4 15 point point VERB fcis-4086 4 16 out out ADP fcis-4086 4 17 the the DET fcis-4086 4 18 current current ADJ fcis-4086 4 19 problems problem NOUN fcis-4086 4 20 of of ADP fcis-4086 4 21 text text NOUN fcis-4086 4 22 classification classification NOUN fcis-4086 4 23 based base VERB fcis-4086 4 24 on on ADP fcis-4086 4 25 traditional traditional ADJ fcis-4086 4 26 machine machine NOUN fcis-4086 4 27 learning learning NOUN fcis-4086 4 28 , , PUNCT fcis-4086 4 29 and and CCONJ fcis-4086 4 30 then then ADV fcis-4086 4 31 describe describe VERB fcis-4086 4 32 the the DET fcis-4086 4 33 superiority superiority NOUN fcis-4086 4 34 of of ADP fcis-4086 4 35 deep deep ADJ fcis-4086 4 36 learning learning NOUN fcis-4086 4 37 in in ADP fcis-4086 4 38 this this DET fcis-4086 4 39 aspect aspect NOUN fcis-4086 4 40 of of ADP fcis-4086 4 41 feature feature NOUN fcis-4086 4 42 extraction extraction NOUN fcis-4086 4 43 . . PUNCT fcis-4086 5 1 after after ADP fcis-4086 5 2 that that PRON fcis-4086 5 3 , , PUNCT fcis-4086 5 4 three three NUM fcis-4086 5 5 different different ADJ fcis-4086 5 6 deep deep ADJ fcis-4086 5 7 learning learning NOUN fcis-4086 5 8 models model NOUN fcis-4086 5 9 are be AUX fcis-4086 5 10 experimented experiment VERB fcis-4086 5 11 , , PUNCT fcis-4086 5 12 and and CCONJ fcis-4086 5 13 the the DET fcis-4086 5 14 differences difference NOUN fcis-4086 5 15 between between ADP fcis-4086 5 16 the the DET fcis-4086 5 17 trained train VERB fcis-4086 5 18 models model NOUN fcis-4086 5 19 are be AUX fcis-4086 5 20 compared compare VERB fcis-4086 5 21 and and CCONJ fcis-4086 5 22 the the DET fcis-4086 5 23 relative relative ADJ fcis-4086 5 24 optimal optimal ADJ fcis-4086 5 25 models model NOUN fcis-4086 5 26 are be AUX fcis-4086 5 27 derived derive VERB fcis-4086 5 28 . . PUNCT fcis-4086 6 1 finally finally ADV fcis-4086 6 2 , , PUNCT fcis-4086 6 3 the the DET fcis-4086 6 4 future future ADJ fcis-4086 6 5 trends trend NOUN fcis-4086 6 6 and and CCONJ fcis-4086 6 7 research research NOUN fcis-4086 6 8 directions direction NOUN fcis-4086 6 9 of of ADP fcis-4086 6 10 deep deep ADJ fcis-4086 6 11 learning learning NOUN fcis-4086 6 12 in in ADP fcis-4086 6 13 this this DET fcis-4086 6 14 field field NOUN fcis-4086 6 15 are be AUX fcis-4086 6 16 discussed discuss VERB fcis-4086 6 17 . . PUNCT fcis-4086 7 1 keywords keyword NOUN fcis-4086 7 2 : : PUNCT fcis-4086 7 3 text text NOUN fcis-4086 7 4 classification classification NOUN fcis-4086 7 5 ; ; PUNCT fcis-4086 7 6 deep deep ADJ fcis-4086 7 7 learning learning NOUN fcis-4086 7 8 ; ; PUNCT fcis-4086 7 9 neural neural ADJ fcis-4086 7 10 networks network NOUN fcis-4086 7 11 . . PUNCT fcis-4086 8 1 1 1 X fcis-4086 8 2 . . X fcis-4086 8 3 background background NOUN fcis-4086 8 4 machine machine NOUN fcis-4086 8 5 learning learning NOUN fcis-4086 8 6 is be AUX fcis-4086 8 7 an an DET fcis-4086 8 8 interdisciplinary interdisciplinary ADJ fcis-4086 8 9 discipline discipline NOUN fcis-4086 8 10 , , PUNCT fcis-4086 8 11 mainly mainly ADV fcis-4086 8 12 applied apply VERB fcis-4086 8 13 to to ADP fcis-4086 8 14 natural natural ADJ fcis-4086 8 15 language language NOUN fcis-4086 8 16 processing processing NOUN fcis-4086 8 17 and and CCONJ fcis-4086 8 18 image image NOUN fcis-4086 8 19 processing processing NOUN fcis-4086 8 20 . . PUNCT fcis-4086 9 1 text text NOUN fcis-4086 9 2 classification classification NOUN fcis-4086 9 3 is be AUX fcis-4086 9 4 a a DET fcis-4086 9 5 branch branch NOUN fcis-4086 9 6 of of ADP fcis-4086 9 7 natural natural ADJ fcis-4086 9 8 language language NOUN fcis-4086 9 9 processing processing NOUN fcis-4086 9 10 , , PUNCT fcis-4086 9 11 and and CCONJ fcis-4086 9 12 its its PRON fcis-4086 9 13 purpose purpose NOUN fcis-4086 9 14 is be AUX fcis-4086 9 15 to to PART fcis-4086 9 16 organize organize VERB fcis-4086 9 17 and and CCONJ fcis-4086 9 18 classify classify VERB fcis-4086 9 19 text text NOUN fcis-4086 9 20 data datum NOUN fcis-4086 9 21 . . PUNCT fcis-4086 10 1 in in ADP fcis-4086 10 2 traditional traditional ADJ fcis-4086 10 3 machine machine NOUN fcis-4086 10 4 learning learning NOUN fcis-4086 10 5 - - PUNCT fcis-4086 10 6 based base VERB fcis-4086 10 7 text text NOUN fcis-4086 10 8 classification classification NOUN fcis-4086 10 9 , , PUNCT fcis-4086 10 10 the the DET fcis-4086 10 11 main main ADJ fcis-4086 10 12 algorithms algorithm NOUN fcis-4086 10 13 are be AUX fcis-4086 10 14 support support NOUN fcis-4086 10 15 vector vector NOUN fcis-4086 10 16 machines machine NOUN fcis-4086 10 17 and and CCONJ fcis-4086 10 18 k k NOUN fcis-4086 10 19 - - PUNCT fcis-4086 10 20 nearest near ADJ fcis-4086 10 21 neighbor neighbor NOUN fcis-4086 10 22 algorithms algorithm NOUN fcis-4086 10 23 , , PUNCT fcis-4086 10 24 etc etc X fcis-4086 10 25 . . X fcis-4086 11 1 however however ADV fcis-4086 11 2 , , PUNCT fcis-4086 11 3 machine machine NOUN fcis-4086 11 4 learning learning NOUN fcis-4086 11 5 - - PUNCT fcis-4086 11 6 based base VERB fcis-4086 11 7 text text NOUN fcis-4086 11 8 classification classification NOUN fcis-4086 11 9 is be AUX fcis-4086 11 10 a a DET fcis-4086 11 11 very very ADV fcis-4086 11 12 important important ADJ fcis-4086 11 13 part part NOUN fcis-4086 11 14 of of ADP fcis-4086 11 15 natural natural ADJ fcis-4086 11 16 language language NOUN fcis-4086 11 17 processing processing NOUN fcis-4086 11 18 . . PUNCT fcis-4086 12 1 however however ADV fcis-4086 12 2 , , PUNCT fcis-4086 12 3 the the DET fcis-4086 12 4 machine machine NOUN fcis-4086 12 5 learning learning NOUN fcis-4086 12 6 - - PUNCT fcis-4086 12 7 based base VERB fcis-4086 12 8 text text NOUN fcis-4086 12 9 classification classification NOUN fcis-4086 12 10 algorithms algorithm NOUN fcis-4086 12 11 ignore ignore VERB fcis-4086 12 12 the the DET fcis-4086 12 13 relationship relationship NOUN fcis-4086 12 14 between between ADP fcis-4086 12 15 word word NOUN fcis-4086 12 16 vectors vector NOUN fcis-4086 12 17 and and CCONJ fcis-4086 12 18 sentences sentence NOUN fcis-4086 12 19 , , PUNCT fcis-4086 12 20 and and CCONJ fcis-4086 12 21 have have VERB fcis-4086 12 22 poor poor ADJ fcis-4086 12 23 processing processing NOUN fcis-4086 12 24 and and CCONJ fcis-4086 12 25 generalization generalization NOUN fcis-4086 12 26 ability ability NOUN fcis-4086 12 27 for for ADP fcis-4086 12 28 high high ADJ fcis-4086 12 29 - - PUNCT fcis-4086 12 30 dimensional dimensional ADJ fcis-4086 12 31 data datum NOUN fcis-4086 12 32 , , PUNCT fcis-4086 12 33 while while SCONJ fcis-4086 12 34 this this DET fcis-4086 12 35 limitation limitation NOUN fcis-4086 12 36 is be AUX fcis-4086 12 37 solved solve VERB fcis-4086 12 38 accordingly accordingly ADV fcis-4086 12 39 after after ADP fcis-4086 12 40 the the DET fcis-4086 12 41 introduction introduction NOUN fcis-4086 12 42 of of ADP fcis-4086 12 43 deep deep ADJ fcis-4086 12 44 learning learning NOUN fcis-4086 12 45 models model NOUN fcis-4086 12 46 . . PUNCT fcis-4086 13 1 the the DET fcis-4086 13 2 essence essence NOUN fcis-4086 13 3 of of ADP fcis-4086 13 4 deep deep ADJ fcis-4086 13 5 learning learning NOUN fcis-4086 13 6 is be AUX fcis-4086 13 7 multilayer multilayer ADJ fcis-4086 13 8 neural neural ADJ fcis-4086 13 9 network network NOUN fcis-4086 13 10 , , PUNCT fcis-4086 13 11 and and CCONJ fcis-4086 13 12 using use VERB fcis-4086 13 13 multilayer multilayer ADJ fcis-4086 13 14 neural neural ADJ fcis-4086 13 15 network network NOUN fcis-4086 13 16 and and CCONJ fcis-4086 13 17 combining combine VERB fcis-4086 13 18 with with ADP fcis-4086 13 19 traditional traditional ADJ fcis-4086 13 20 machine machine NOUN fcis-4086 13 21 learning learning NOUN fcis-4086 13 22 algorithm algorithm NOUN fcis-4086 13 23 , , PUNCT fcis-4086 13 24 feature feature NOUN fcis-4086 13 25 extraction extraction NOUN fcis-4086 13 26 of of ADP fcis-4086 13 27 large large ADJ fcis-4086 13 28 amount amount NOUN fcis-4086 13 29 of of ADP fcis-4086 13 30 data datum NOUN fcis-4086 13 31 can can AUX fcis-4086 13 32 be be AUX fcis-4086 13 33 done do VERB fcis-4086 13 34 automatically automatically ADV fcis-4086 13 35 by by ADP fcis-4086 13 36 computer computer NOUN fcis-4086 13 37 . . PUNCT fcis-4086 14 1 throughout throughout ADP fcis-4086 14 2 the the DET fcis-4086 14 3 training training NOUN fcis-4086 14 4 process process NOUN fcis-4086 14 5 , , PUNCT fcis-4086 14 6 model model NOUN fcis-4086 14 7 learning learning NOUN fcis-4086 14 8 is be AUX fcis-4086 14 9 automated automate VERB fcis-4086 14 10 , , PUNCT fcis-4086 14 11 the the DET fcis-4086 14 12 design design NOUN fcis-4086 14 13 cost cost NOUN fcis-4086 14 14 of of ADP fcis-4086 14 15 the the DET fcis-4086 14 16 problem problem NOUN fcis-4086 14 17 is be AUX fcis-4086 14 18 reduced reduce VERB fcis-4086 14 19 , , PUNCT fcis-4086 14 20 and and CCONJ fcis-4086 14 21 the the DET fcis-4086 14 22 data datum NOUN fcis-4086 14 23 information information NOUN fcis-4086 14 24 analysis analysis NOUN fcis-4086 14 25 and and CCONJ fcis-4086 14 26 extraction extraction NOUN fcis-4086 14 27 capability capability NOUN fcis-4086 14 28 is be AUX fcis-4086 14 29 enhanced enhance VERB fcis-4086 14 30 . . PUNCT fcis-4086 15 1 the the DET fcis-4086 15 2 mainstream mainstream NOUN fcis-4086 15 3 deep deep ADJ fcis-4086 15 4 learning learning NOUN fcis-4086 15 5 models model NOUN fcis-4086 15 6 are be AUX fcis-4086 15 7 convolutional convolutional ADJ fcis-4086 15 8 neural neural ADJ fcis-4086 15 9 networks network NOUN fcis-4086 15 10 ( ( PUNCT fcis-4086 15 11 cnn cnn PROPN fcis-4086 15 12 ) ) PUNCT fcis-4086 15 13 , , PUNCT fcis-4086 15 14 recurrent recurrent ADJ fcis-4086 15 15 neural neural ADJ fcis-4086 15 16 networks network NOUN fcis-4086 15 17 ( ( PUNCT fcis-4086 15 18 rnn rnn PROPN fcis-4086 15 19 ) ) PUNCT fcis-4086 15 20 , , PUNCT fcis-4086 15 21 attention attention NOUN fcis-4086 15 22 mechanisms mechanism NOUN fcis-4086 15 23 and and CCONJ fcis-4086 15 24 other other ADJ fcis-4086 15 25 models model NOUN fcis-4086 15 26 . . PUNCT fcis-4086 16 1 these these DET fcis-4086 16 2 models model NOUN fcis-4086 16 3 have have AUX fcis-4086 16 4 achieved achieve VERB fcis-4086 16 5 good good ADJ fcis-4086 16 6 experimental experimental ADJ fcis-4086 16 7 results result NOUN fcis-4086 16 8 in in ADP fcis-4086 16 9 the the DET fcis-4086 16 10 field field NOUN fcis-4086 16 11 of of ADP fcis-4086 16 12 text text NOUN fcis-4086 16 13 classification classification NOUN fcis-4086 16 14 . . PUNCT fcis-4086 17 1 in in ADP fcis-4086 17 2 this this DET fcis-4086 17 3 paper paper NOUN fcis-4086 17 4 , , PUNCT fcis-4086 17 5 we we PRON fcis-4086 17 6 will will AUX fcis-4086 17 7 first first ADV fcis-4086 17 8 give give VERB fcis-4086 17 9 a a DET fcis-4086 17 10 detailed detailed ADJ fcis-4086 17 11 introduction introduction NOUN fcis-4086 17 12 to to ADP fcis-4086 17 13 the the DET fcis-4086 17 14 traditional traditional ADJ fcis-4086 17 15 text text NOUN fcis-4086 17 16 classification classification NOUN fcis-4086 17 17 process process NOUN fcis-4086 17 18 and and CCONJ fcis-4086 17 19 show show VERB fcis-4086 17 20 the the DET fcis-4086 17 21 problems problem NOUN fcis-4086 17 22 of of ADP fcis-4086 17 23 traditional traditional ADJ fcis-4086 17 24 methods method NOUN fcis-4086 17 25 , , PUNCT fcis-4086 17 26 then then ADV fcis-4086 17 27 describe describe VERB fcis-4086 17 28 three three NUM fcis-4086 17 29 commonly commonly ADV fcis-4086 17 30 used use VERB fcis-4086 17 31 deep deep ADJ fcis-4086 17 32 learning learning NOUN fcis-4086 17 33 models model NOUN fcis-4086 17 34 and and CCONJ fcis-4086 17 35 conduct conduct VERB fcis-4086 17 36 experiments experiment NOUN fcis-4086 17 37 with with ADP fcis-4086 17 38 them they PRON fcis-4086 17 39 , , PUNCT fcis-4086 17 40 and and CCONJ fcis-4086 17 41 finally finally ADV fcis-4086 17 42 compare compare VERB fcis-4086 17 43 the the DET fcis-4086 17 44 optimal optimal ADJ fcis-4086 17 45 model model NOUN fcis-4086 17 46 among among ADP fcis-4086 17 47 the the DET fcis-4086 17 48 three three NUM fcis-4086 17 49 models model NOUN fcis-4086 17 50 by by ADP fcis-4086 17 51 experimental experimental ADJ fcis-4086 17 52 data datum NOUN fcis-4086 17 53 . . PUNCT fcis-4086 18 1 2 2 X fcis-4086 18 2 . . X fcis-4086 18 3 methods method NOUN fcis-4086 18 4 and and CCONJ fcis-4086 18 5 models model NOUN fcis-4086 18 6 there there PRON fcis-4086 18 7 are be VERB fcis-4086 18 8 two two NUM fcis-4086 18 9 main main ADJ fcis-4086 18 10 types type NOUN fcis-4086 18 11 of of ADP fcis-4086 18 12 text text NOUN fcis-4086 18 13 classification classification NOUN fcis-4086 18 14 methods method NOUN fcis-4086 18 15 : : PUNCT fcis-4086 18 16 traditional traditional ADJ fcis-4086 18 17 text text NOUN fcis-4086 18 18 classification classification NOUN fcis-4086 18 19 ; ; PUNCT fcis-4086 18 20 deep deep ADJ fcis-4086 18 21 learning learn VERB fcis-4086 18 22 text text NOUN fcis-4086 18 23 classification classification NOUN fcis-4086 18 24 . . PUNCT fcis-4086 19 1 for for ADP fcis-4086 19 2 the the DET fcis-4086 19 3 former former ADJ fcis-4086 19 4 , , PUNCT fcis-4086 19 5 traditional traditional ADJ fcis-4086 19 6 text text NOUN fcis-4086 19 7 classification classification NOUN fcis-4086 19 8 should should AUX fcis-4086 19 9 first first ADV fcis-4086 19 10 select select VERB fcis-4086 19 11 a a DET fcis-4086 19 12 dataset dataset NOUN fcis-4086 19 13 , , PUNCT fcis-4086 19 14 and and CCONJ fcis-4086 19 15 then then ADV fcis-4086 19 16 manually manually ADV fcis-4086 19 17 construct construct VERB fcis-4086 19 18 feature feature NOUN fcis-4086 19 19 engineering engineering NOUN fcis-4086 19 20 , , PUNCT fcis-4086 19 21 which which PRON fcis-4086 19 22 contains contain VERB fcis-4086 19 23 several several ADJ fcis-4086 19 24 major major ADJ fcis-4086 19 25 steps step NOUN fcis-4086 19 26 , , PUNCT fcis-4086 19 27 including include VERB fcis-4086 19 28 text text NOUN fcis-4086 19 29 pre pre ADJ fcis-4086 19 30 - - ADJ fcis-4086 19 31 processing processing ADJ fcis-4086 19 32 , , PUNCT fcis-4086 19 33 feature feature NOUN fcis-4086 19 34 extraction extraction NOUN fcis-4086 19 35 , , PUNCT fcis-4086 19 36 and and CCONJ fcis-4086 19 37 feature feature NOUN fcis-4086 19 38 representation representation NOUN fcis-4086 19 39 . . PUNCT fcis-4086 20 1 finally finally ADV fcis-4086 20 2 , , PUNCT fcis-4086 20 3 a a DET fcis-4086 20 4 suitable suitable ADJ fcis-4086 20 5 machine machine NOUN fcis-4086 20 6 learning learn VERB fcis-4086 20 7 algorithm algorithm NOUN fcis-4086 20 8 is be AUX fcis-4086 20 9 selected select VERB fcis-4086 20 10 to to PART fcis-4086 20 11 classify classify VERB fcis-4086 20 12 the the DET fcis-4086 20 13 text text NOUN fcis-4086 20 14 after after ADP fcis-4086 20 15 the the DET fcis-4086 20 16 feature feature NOUN fcis-4086 20 17 engineering engineering NOUN fcis-4086 20 18 process process NOUN fcis-4086 20 19 . . PUNCT fcis-4086 21 1 this this DET fcis-4086 21 2 process process NOUN fcis-4086 21 3 is be AUX fcis-4086 21 4 described describe VERB fcis-4086 21 5 in in ADP fcis-4086 21 6 detail detail NOUN fcis-4086 21 7 below below ADV fcis-4086 21 8 . . PUNCT fcis-4086 22 1 figure figure NOUN fcis-4086 22 2 1 1 NUM fcis-4086 22 3 . . PUNCT fcis-4086 23 1 structure structure NOUN fcis-4086 23 2 of of ADP fcis-4086 23 3 news news NOUN fcis-4086 23 4 text text NOUN fcis-4086 23 5 classification classification NOUN fcis-4086 23 6 algorithm algorithm NOUN fcis-4086 23 7 2.1 2.1 NUM fcis-4086 23 8 . . PUNCT fcis-4086 24 1 text text NOUN fcis-4086 24 2 pre pre ADJ fcis-4086 24 3 - - ADJ fcis-4086 24 4 processing process VERB fcis-4086 24 5 the the DET fcis-4086 24 6 core core NOUN fcis-4086 24 7 steps step NOUN fcis-4086 24 8 of of ADP fcis-4086 24 9 text text NOUN fcis-4086 24 10 pre pre ADJ fcis-4086 24 11 - - ADJ fcis-4086 24 12 processing processing NOUN fcis-4086 24 13 are be AUX fcis-4086 24 14 text text NOUN fcis-4086 24 15 splitting splitting NOUN fcis-4086 24 16 and and CCONJ fcis-4086 24 17 text text NOUN fcis-4086 24 18 cleaning cleaning NOUN fcis-4086 24 19 . . PUNCT fcis-4086 25 1 firstly firstly ADV fcis-4086 25 2 , , PUNCT fcis-4086 25 3 the the DET fcis-4086 25 4 original original ADJ fcis-4086 25 5 text text NOUN fcis-4086 25 6 data datum NOUN fcis-4086 25 7 collection collection NOUN fcis-4086 25 8 is be AUX fcis-4086 25 9 divided divide VERB fcis-4086 25 10 into into ADP fcis-4086 25 11 a a DET fcis-4086 25 12 single single ADJ fcis-4086 25 13 chapter chapter NOUN fcis-4086 25 14 - - PUNCT fcis-4086 25 15 based base VERB fcis-4086 25 16 collection collection NOUN fcis-4086 25 17 , , PUNCT fcis-4086 25 18 and and CCONJ fcis-4086 25 19 then then ADV fcis-4086 25 20 the the DET fcis-4086 25 21 text text NOUN fcis-4086 25 22 is be AUX fcis-4086 25 23 divided divide VERB fcis-4086 25 24 into into ADP fcis-4086 25 25 words word NOUN fcis-4086 25 26 according accord VERB fcis-4086 25 27 to to ADP fcis-4086 25 28 certain certain ADJ fcis-4086 25 29 rules rule NOUN fcis-4086 25 30 for for ADP fcis-4086 25 31 extracting extract VERB fcis-4086 25 32 the the DET fcis-4086 25 33 feature feature NOUN fcis-4086 25 34 values value NOUN fcis-4086 25 35 of of ADP fcis-4086 25 36 the the DET fcis-4086 25 37 text text NOUN fcis-4086 25 38 through through ADP fcis-4086 25 39 the the DET fcis-4086 25 40 text text NOUN fcis-4086 25 41 splitting splitting NOUN fcis-4086 25 42 process process NOUN fcis-4086 25 43 . . PUNCT fcis-4086 26 1 the the DET fcis-4086 26 2 dictionary dictionary NOUN fcis-4086 26 3 is be AUX fcis-4086 26 4 constructed construct VERB fcis-4086 26 5 first first ADV fcis-4086 26 6 , , PUNCT fcis-4086 26 7 and and CCONJ fcis-4086 26 8 then then ADV fcis-4086 26 9 the the DET fcis-4086 26 10 text text NOUN fcis-4086 26 11 is be AUX fcis-4086 26 12 divided divide VERB fcis-4086 26 13 into into ADP fcis-4086 26 14 words word NOUN fcis-4086 26 15 using use VERB fcis-4086 26 16 a a DET fcis-4086 26 17 dictionary dictionary ADJ fcis-4086 26 18 algorithm algorithm NOUN fcis-4086 26 19 . . PUNCT fcis-4086 27 1 the the DET fcis-4086 27 2 current current ADJ fcis-4086 27 3 popular popular ADJ fcis-4086 27 4 method method NOUN fcis-4086 27 5 for for ADP fcis-4086 27 6 constructing construct VERB fcis-4086 27 7 dictionaries dictionary NOUN fcis-4086 27 8 is be AUX fcis-4086 27 9 dictionary dictionary ADJ fcis-4086 27 10 trees tree NOUN fcis-4086 27 11 , , PUNCT fcis-4086 27 12 and and CCONJ fcis-4086 27 13 the the DET fcis-4086 27 14 word word NOUN fcis-4086 27 15 separation separation NOUN fcis-4086 27 16 algorithms algorithm NOUN fcis-4086 27 17 can can AUX fcis-4086 27 18 be be AUX fcis-4086 27 19 divided divide VERB fcis-4086 27 20 into into ADP fcis-4086 27 21 three three NUM fcis-4086 27 22 categories category NOUN fcis-4086 27 23 : : PUNCT fcis-4086 27 24 string string NOUN fcis-4086 27 25 matching matching NOUN fcis-4086 27 26 - - PUNCT fcis-4086 27 27 based base VERB fcis-4086 27 28 word word NOUN fcis-4086 27 29 separation separation NOUN fcis-4086 27 30 algorithms algorithm NOUN fcis-4086 27 31 , , PUNCT fcis-4086 27 32 such such ADJ fcis-4086 27 33 as as ADP fcis-4086 27 34 forward forward ADV fcis-4086 27 35 maximal maximal ADJ fcis-4086 27 36 matching matching NOUN fcis-4086 27 37 algorithms algorithm NOUN fcis-4086 27 38 ; ; PUNCT fcis-4086 27 39 statistical statistical ADJ fcis-4086 27 40 - - PUNCT fcis-4086 27 41 based base VERB fcis-4086 27 42 word word NOUN fcis-4086 27 43 separation separation NOUN fcis-4086 27 44 algorithms algorithm NOUN fcis-4086 27 45 such such ADJ fcis-4086 27 46 as as ADP fcis-4086 27 47 crf crf PROPN fcis-4086 27 48 and and CCONJ fcis-4086 27 49 n n CCONJ fcis-4086 27 50 - - PUNCT fcis-4086 27 51 gram gram NOUN fcis-4086 27 52 ( ( PUNCT fcis-4086 27 53 n n CCONJ fcis-4086 27 54 - - PUNCT fcis-4086 27 55 gram gram NOUN fcis-4086 27 56 ) ) PUNCT fcis-4086 27 57 ; ; PUNCT fcis-4086 27 58 and and CCONJ fcis-4086 27 59 understanding understanding NOUN fcis-4086 27 60 - - PUNCT fcis-4086 27 61 based base VERB fcis-4086 27 62 word word NOUN fcis-4086 27 63 separation separation NOUN fcis-4086 27 64 methods method NOUN fcis-4086 27 65 , , PUNCT fcis-4086 27 66 i.e. i.e. X fcis-4086 27 67 , , PUNCT fcis-4086 27 68 artificial artificial ADJ fcis-4086 27 69 intelligence intelligence NOUN fcis-4086 27 70 methods method NOUN fcis-4086 27 71 . . PUNCT fcis-4086 28 1 finally finally ADV fcis-4086 28 2 , , PUNCT fcis-4086 28 3 text text NOUN fcis-4086 28 4 cleaning cleaning NOUN fcis-4086 28 5 , , PUNCT fcis-4086 28 6 i.e. i.e. X fcis-4086 28 7 , , PUNCT fcis-4086 28 8 removing remove VERB fcis-4086 28 9 noise noise NOUN fcis-4086 28 10 from from ADP fcis-4086 28 11 the the DET fcis-4086 28 12 data datum NOUN fcis-4086 28 13 , , PUNCT fcis-4086 28 14 such such ADJ fcis-4086 28 15 as as ADP fcis-4086 28 16 discontinued discontinued ADJ fcis-4086 28 17 words word NOUN fcis-4086 28 18 , , PUNCT fcis-4086 28 19 special special ADJ fcis-4086 28 20 symbols symbol NOUN fcis-4086 28 21 , , PUNCT fcis-4086 28 22 and and CCONJ fcis-4086 28 23 those those DET fcis-4086 28 24 words word NOUN fcis-4086 28 25 or or CCONJ fcis-4086 28 26 characters character NOUN fcis-4086 28 27 that that PRON fcis-4086 28 28 do do AUX fcis-4086 28 29 not not PART fcis-4086 28 30 characterize characterize VERB fcis-4086 28 31 this this DET fcis-4086 28 32 text text NOUN fcis-4086 28 33 . . PUNCT fcis-4086 29 1 the the DET fcis-4086 29 2 idea idea NOUN fcis-4086 29 3 of of ADP fcis-4086 29 4 the the DET fcis-4086 29 5 forward forward ADV fcis-4086 29 6 maximum maximum ADJ fcis-4086 29 7 matching matching NOUN fcis-4086 29 8 algorithm algorithm NOUN fcis-4086 29 9 is be AUX fcis-4086 29 10 as as SCONJ fcis-4086 29 11 follows follow VERB fcis-4086 29 12 : : PUNCT fcis-4086 29 13 initialize initialize VERB fcis-4086 29 14 the the DET fcis-4086 29 15 word word NOUN fcis-4086 29 16 string string NOUN fcis-4086 29 17 s1 s1 NOUN fcis-4086 29 18 to to PART fcis-4086 29 19 be be AUX fcis-4086 29 20 cut cut VERB fcis-4086 29 21 , , PUNCT fcis-4086 29 22 output output VERB fcis-4086 29 23 the the DET fcis-4086 29 24 word word NOUN fcis-4086 29 25 string string PROPN fcis-4086 29 26 s2 s2 PROPN fcis-4086 29 27 , , PUNCT fcis-4086 29 28 maximum maximum ADJ fcis-4086 29 29 length length NOUN fcis-4086 29 30 m m PROPN fcis-4086 29 31 , , PUNCT fcis-4086 29 32 then then ADV fcis-4086 29 33 : : PUNCT fcis-4086 29 34 while(s1 while(s1 NOUN fcis-4086 29 35 is be AUX fcis-4086 29 36 not not PART fcis-4086 29 37 empty empty ADJ fcis-4086 29 38 ) ) PUNCT fcis-4086 29 39 { { PUNCT fcis-4086 29 40 fetch fetch VERB fcis-4086 29 41 the the DET fcis-4086 29 42 candidate candidate NOUN fcis-4086 29 43 string string NOUN fcis-4086 29 44 w w ADP fcis-4086 29 45 whose whose DET fcis-4086 29 46 length length NOUN fcis-4086 29 47 is be AUX fcis-4086 29 48 less less ADJ fcis-4086 29 49 than than ADP fcis-4086 29 50 m. m. NOUN fcis-4086 29 51 query query NOUN fcis-4086 29 52 whether whether SCONJ fcis-4086 29 53 w w NOUN fcis-4086 29 54 is be AUX fcis-4086 29 55 in in ADP fcis-4086 29 56 the the DET fcis-4086 29 57 dictionary dictionary PROPN fcis-4086 29 58 . . PUNCT fcis-4086 30 1 remove remove VERB fcis-4086 30 2 the the DET fcis-4086 30 3 rightmost rightmost ADJ fcis-4086 30 4 word word NOUN fcis-4086 30 5 of of ADP fcis-4086 30 6 w. w. PROPN fcis-4086 30 7 if(w if(w PUNCT fcis-4086 30 8 is be AUX fcis-4086 30 9 a a DET fcis-4086 30 10 single single ADJ fcis-4086 30 11 word word NOUN fcis-4086 30 12 ) ) PUNCT fcis-4086 30 13 { { PUNCT fcis-4086 30 14 59 59 NUM fcis-4086 30 15 s2 s2 NOUN fcis-4086 30 16 = = SYM fcis-4086 30 17 s2 s2 PROPN fcis-4086 30 18 + + CCONJ fcis-4086 30 19 w w NOUN fcis-4086 30 20 s1 s1 NOUN fcis-4086 30 21 = = SYM fcis-4086 30 22 s1 s1 PROPN fcis-4086 30 23 βˆ’ βˆ’ PROPN fcis-4086 30 24 w w NOUN fcis-4086 30 25 } } PUNCT fcis-4086 30 26 } } PUNCT fcis-4086 30 27 2.2 2.2 NUM fcis-4086 30 28 . . PUNCT fcis-4086 31 1 feature feature NOUN fcis-4086 31 2 extraction extraction NOUN fcis-4086 31 3 the the DET fcis-4086 31 4 purpose purpose NOUN fcis-4086 31 5 of of ADP fcis-4086 31 6 feature feature NOUN fcis-4086 31 7 extraction extraction NOUN fcis-4086 31 8 is be AUX fcis-4086 31 9 to to PART fcis-4086 31 10 reduce reduce VERB fcis-4086 31 11 the the DET fcis-4086 31 12 number number NOUN fcis-4086 31 13 of of ADP fcis-4086 31 14 words word NOUN fcis-4086 31 15 to to PART fcis-4086 31 16 be be AUX fcis-4086 31 17 processed process VERB fcis-4086 31 18 by by ADP fcis-4086 31 19 the the DET fcis-4086 31 20 model model NOUN fcis-4086 31 21 as as ADV fcis-4086 31 22 much much ADV fcis-4086 31 23 as as ADP fcis-4086 31 24 possible possible ADJ fcis-4086 31 25 without without ADP fcis-4086 31 26 damaging damage VERB fcis-4086 31 27 the the DET fcis-4086 31 28 core core ADJ fcis-4086 31 29 information information NOUN fcis-4086 31 30 of of ADP fcis-4086 31 31 the the DET fcis-4086 31 32 text text NOUN fcis-4086 31 33 , , PUNCT fcis-4086 31 34 so so SCONJ fcis-4086 31 35 as as SCONJ fcis-4086 31 36 to to PART fcis-4086 31 37 reduce reduce VERB fcis-4086 31 38 the the DET fcis-4086 31 39 vector vector NOUN fcis-4086 31 40 space space NOUN fcis-4086 31 41 dimension dimension NOUN fcis-4086 31 42 and and CCONJ fcis-4086 31 43 thus thus ADV fcis-4086 31 44 simplify simplify VERB fcis-4086 31 45 the the DET fcis-4086 31 46 operation operation NOUN fcis-4086 31 47 . . PUNCT fcis-4086 32 1 2.2.1 2.2.1 NUM fcis-4086 32 2 . . PUNCT fcis-4086 32 3 machine machine NOUN fcis-4086 32 4 learning learning NOUN fcis-4086 32 5 based base VERB fcis-4086 32 6 feature feature NOUN fcis-4086 32 7 extraction extraction NOUN fcis-4086 32 8 feature feature NOUN fcis-4086 32 9 extraction extraction NOUN fcis-4086 32 10 includes include VERB fcis-4086 32 11 feature feature NOUN fcis-4086 32 12 selection selection NOUN fcis-4086 32 13 and and CCONJ fcis-4086 32 14 feature feature NOUN fcis-4086 32 15 weight weight NOUN fcis-4086 32 16 calculation calculation NOUN fcis-4086 32 17 . . PUNCT fcis-4086 33 1 the the DET fcis-4086 33 2 basic basic ADJ fcis-4086 33 3 idea idea NOUN fcis-4086 33 4 of of ADP fcis-4086 33 5 feature feature NOUN fcis-4086 33 6 selection selection NOUN fcis-4086 33 7 is be AUX fcis-4086 33 8 to to PART fcis-4086 33 9 rank rank VERB fcis-4086 33 10 the the DET fcis-4086 33 11 original original ADJ fcis-4086 33 12 feature feature NOUN fcis-4086 33 13 items item NOUN fcis-4086 33 14 according accord VERB fcis-4086 33 15 to to ADP fcis-4086 33 16 a a DET fcis-4086 33 17 certain certain ADJ fcis-4086 33 18 evaluation evaluation NOUN fcis-4086 33 19 index index NOUN fcis-4086 33 20 , , PUNCT fcis-4086 33 21 select select VERB fcis-4086 33 22 the the DET fcis-4086 33 23 highest high ADJ fcis-4086 33 24 scoring scoring NOUN fcis-4086 33 25 feature feature NOUN fcis-4086 33 26 items item NOUN fcis-4086 33 27 , , PUNCT fcis-4086 33 28 and and CCONJ fcis-4086 33 29 filter filter VERB fcis-4086 33 30 the the DET fcis-4086 33 31 rest rest NOUN fcis-4086 33 32 . . PUNCT fcis-4086 34 1 for for ADP fcis-4086 34 2 feature feature NOUN fcis-4086 34 3 weighting weighting NOUN fcis-4086 34 4 , , PUNCT fcis-4086 34 5 the the DET fcis-4086 34 6 tf tf PROPN fcis-4086 34 7 - - PUNCT fcis-4086 34 8 idf idf PROPN fcis-4086 34 9 algorithm algorithm NOUN fcis-4086 34 10 is be AUX fcis-4086 34 11 generally generally ADV fcis-4086 34 12 chosen choose VERB fcis-4086 34 13 to to PART fcis-4086 34 14 evaluate evaluate VERB fcis-4086 34 15 the the DET fcis-4086 34 16 importance importance NOUN fcis-4086 34 17 of of ADP fcis-4086 34 18 a a DET fcis-4086 34 19 word word NOUN fcis-4086 34 20 to to ADP fcis-4086 34 21 a a DET fcis-4086 34 22 document document NOUN fcis-4086 34 23 set set NOUN fcis-4086 34 24 . . PUNCT fcis-4086 35 1 when when SCONJ fcis-4086 35 2 the the DET fcis-4086 35 3 word word NOUN fcis-4086 35 4 frequency frequency NOUN fcis-4086 35 5 and and CCONJ fcis-4086 35 6 inverse inverse NOUN fcis-4086 35 7 document document NOUN fcis-4086 35 8 frequency frequency NOUN fcis-4086 35 9 are be AUX fcis-4086 35 10 available available ADJ fcis-4086 35 11 , , PUNCT fcis-4086 35 12 these these DET fcis-4086 35 13 two two NUM fcis-4086 35 14 words word NOUN fcis-4086 35 15 are be AUX fcis-4086 35 16 multiplied multiply VERB fcis-4086 35 17 together together ADV fcis-4086 35 18 to to PART fcis-4086 35 19 obtain obtain VERB fcis-4086 35 20 the the DET fcis-4086 35 21 value value NOUN fcis-4086 35 22 of of ADP fcis-4086 35 23 tf tf PROPN fcis-4086 35 24 - - PUNCT fcis-4086 35 25 idf idf PROPN fcis-4086 35 26 for for ADP fcis-4086 35 27 a a DET fcis-4086 35 28 word word NOUN fcis-4086 35 29 , , PUNCT fcis-4086 35 30 and and CCONJ fcis-4086 35 31 the the PRON fcis-4086 35 32 larger large ADJ fcis-4086 35 33 this this DET fcis-4086 35 34 value value NOUN fcis-4086 35 35 is be AUX fcis-4086 35 36 , , PUNCT fcis-4086 35 37 the the PRON fcis-4086 35 38 greater great ADJ fcis-4086 35 39 the the DET fcis-4086 35 40 probability probability NOUN fcis-4086 35 41 that that SCONJ fcis-4086 35 42 the the DET fcis-4086 35 43 word word NOUN fcis-4086 35 44 will will AUX fcis-4086 35 45 become become VERB fcis-4086 35 46 a a DET fcis-4086 35 47 keyword keyword NOUN fcis-4086 35 48 . . PUNCT fcis-4086 36 1 this this DET fcis-4086 36 2 calculation calculation NOUN fcis-4086 36 3 can can AUX fcis-4086 36 4 effectively effectively ADV fcis-4086 36 5 avoid avoid VERB fcis-4086 36 6 the the DET fcis-4086 36 7 influence influence NOUN fcis-4086 36 8 of of ADP fcis-4086 36 9 common common ADJ fcis-4086 36 10 words word NOUN fcis-4086 36 11 on on ADP fcis-4086 36 12 keywords keyword NOUN fcis-4086 36 13 and and CCONJ fcis-4086 36 14 improve improve VERB fcis-4086 36 15 the the DET fcis-4086 36 16 relevance relevance NOUN fcis-4086 36 17 between between ADP fcis-4086 36 18 keywords keyword NOUN fcis-4086 36 19 and and CCONJ fcis-4086 36 20 articles article NOUN fcis-4086 36 21 . . PUNCT fcis-4086 37 1 𝑑𝑓𝑖𝑗 𝑑𝑓𝑖𝑗 ADJ fcis-4086 37 2 = = SYM fcis-4086 37 3 𝑛𝑖𝑗 𝑛𝑖𝑗 ADJ fcis-4086 37 4 βˆ‘ βˆ‘ PROPN fcis-4086 37 5 π‘›π‘˜π‘—π‘˜ π‘›π‘˜π‘—π‘˜ NOUN fcis-4086 37 6 ( ( PUNCT fcis-4086 37 7 1 1 NUM fcis-4086 37 8 ) ) PUNCT fcis-4086 37 9 𝑖𝑑𝑓𝑖 𝑖𝑑𝑓𝑖 NOUN fcis-4086 37 10 = = SYM fcis-4086 37 11 log log PROPN fcis-4086 37 12 ( ( PUNCT fcis-4086 37 13 |𝐷| |𝐷| X fcis-4086 37 14 1+|𝐷𝑖| 1+|𝐷𝑖| NUM fcis-4086 37 15 ) ) PUNCT fcis-4086 37 16 ( ( PUNCT fcis-4086 37 17 2 2 X fcis-4086 37 18 ) ) PUNCT fcis-4086 37 19 𝑑𝑓 𝑑𝑓 ADP fcis-4086 37 20 βˆ— βˆ— NOUN fcis-4086 37 21 𝑖𝑑𝑓(𝑖 𝑖𝑑𝑓(𝑖 PROPN fcis-4086 37 22 , , PUNCT fcis-4086 37 23 𝑗 𝑗 NOUN fcis-4086 37 24 ) ) PUNCT fcis-4086 37 25 = = SYM fcis-4086 37 26 𝑑𝑓𝑖𝑗 𝑑𝑓𝑖𝑗 ADJ fcis-4086 37 27 βˆ— βˆ— NOUN fcis-4086 37 28 𝑖𝑑𝑓𝑖 𝑖𝑑𝑓𝑖 NOUN fcis-4086 37 29 = = PRON fcis-4086 37 30 𝑛𝑖𝑗 𝑛𝑖𝑗 ADJ fcis-4086 37 31 βˆ‘ βˆ‘ PROPN fcis-4086 37 32 π‘›π‘˜π‘—π‘˜ π‘›π‘˜π‘—π‘˜ PROPN fcis-4086 37 33 βˆ— βˆ— PROPN fcis-4086 37 34 log log PROPN fcis-4086 37 35 ( ( PUNCT fcis-4086 37 36 |𝐷| |𝐷| X fcis-4086 37 37 1+|𝐷𝑖| 1+|𝐷𝑖| NUM fcis-4086 37 38 ) ) PUNCT fcis-4086 37 39 ( ( PUNCT fcis-4086 37 40 3 3 X fcis-4086 37 41 ) ) PUNCT fcis-4086 37 42 2.2.2 2.2.2 NUM fcis-4086 37 43 . . PUNCT fcis-4086 38 1 deep deep ADJ fcis-4086 38 2 learning learning NOUN fcis-4086 38 3 - - PUNCT fcis-4086 38 4 based base VERB fcis-4086 38 5 feature feature NOUN fcis-4086 38 6 extraction extraction NOUN fcis-4086 38 7 the the DET fcis-4086 38 8 difference difference NOUN fcis-4086 38 9 between between ADP fcis-4086 38 10 machine machine NOUN fcis-4086 38 11 learning learning NOUN fcis-4086 38 12 and and CCONJ fcis-4086 38 13 deep deep ADJ fcis-4086 38 14 learning learning NOUN fcis-4086 38 15 is be AUX fcis-4086 38 16 the the DET fcis-4086 38 17 interpretability interpretability NOUN fcis-4086 38 18 of of ADP fcis-4086 38 19 features feature NOUN fcis-4086 38 20 . . PUNCT fcis-4086 39 1 in in ADP fcis-4086 39 2 machine machine NOUN fcis-4086 39 3 learning learning NOUN fcis-4086 39 4 , , PUNCT fcis-4086 39 5 each each DET fcis-4086 39 6 feature feature NOUN fcis-4086 39 7 is be AUX fcis-4086 39 8 required require VERB fcis-4086 39 9 to to PART fcis-4086 39 10 have have VERB fcis-4086 39 11 a a DET fcis-4086 39 12 certain certain ADJ fcis-4086 39 13 meaning meaning NOUN fcis-4086 39 14 as as ADV fcis-4086 39 15 much much ADJ fcis-4086 39 16 as as ADP fcis-4086 39 17 possible possible ADJ fcis-4086 39 18 , , PUNCT fcis-4086 39 19 which which PRON fcis-4086 39 20 is be AUX fcis-4086 39 21 equivalent equivalent ADJ fcis-4086 39 22 to to ADP fcis-4086 39 23 manually manually ADV fcis-4086 39 24 extracting extract VERB fcis-4086 39 25 features feature NOUN fcis-4086 39 26 from from ADP fcis-4086 39 27 the the DET fcis-4086 39 28 original original ADJ fcis-4086 39 29 input input NOUN fcis-4086 39 30 , , PUNCT fcis-4086 39 31 and and CCONJ fcis-4086 39 32 then then ADV fcis-4086 39 33 modeling modeling NOUN fcis-4086 39 34 with with ADP fcis-4086 39 35 the the DET fcis-4086 39 36 extracted extract VERB fcis-4086 39 37 features feature NOUN fcis-4086 39 38 . . PUNCT fcis-4086 40 1 in in ADP fcis-4086 40 2 contrast contrast NOUN fcis-4086 40 3 , , PUNCT fcis-4086 40 4 deep deep ADJ fcis-4086 40 5 learning learning NOUN fcis-4086 40 6 learns learn VERB fcis-4086 40 7 features feature NOUN fcis-4086 40 8 automatically automatically ADV fcis-4086 40 9 from from ADP fcis-4086 40 10 the the DET fcis-4086 40 11 data datum NOUN fcis-4086 40 12 rather rather ADV fcis-4086 40 13 than than ADP fcis-4086 40 14 selecting select VERB fcis-4086 40 15 them they PRON fcis-4086 40 16 manually manually ADV fcis-4086 40 17 . . PUNCT fcis-4086 41 1 for for ADP fcis-4086 41 2 all all DET fcis-4086 41 3 layers layer NOUN fcis-4086 41 4 before before ADP fcis-4086 41 5 the the DET fcis-4086 41 6 output output NOUN fcis-4086 41 7 layer layer NOUN fcis-4086 41 8 , , PUNCT fcis-4086 41 9 it it PRON fcis-4086 41 10 can can AUX fcis-4086 41 11 be be AUX fcis-4086 41 12 viewed view VERB fcis-4086 41 13 as as ADP fcis-4086 41 14 a a DET fcis-4086 41 15 feature feature NOUN fcis-4086 41 16 extraction extraction NOUN fcis-4086 41 17 process process NOUN fcis-4086 41 18 . . PUNCT fcis-4086 42 1 after after ADP fcis-4086 42 2 abstracting abstract VERB fcis-4086 42 3 features feature NOUN fcis-4086 42 4 by by ADP fcis-4086 42 5 passing pass VERB fcis-4086 42 6 the the DET fcis-4086 42 7 original original ADJ fcis-4086 42 8 features feature NOUN fcis-4086 42 9 through through ADP fcis-4086 42 10 a a DET fcis-4086 42 11 multilayer multilayer ADJ fcis-4086 42 12 neural neural ADJ fcis-4086 42 13 network network NOUN fcis-4086 42 14 , , PUNCT fcis-4086 42 15 the the DET fcis-4086 42 16 extracted extract VERB fcis-4086 42 17 features feature NOUN fcis-4086 42 18 are be AUX fcis-4086 42 19 then then ADV fcis-4086 42 20 fed feed VERB fcis-4086 42 21 to to ADP fcis-4086 42 22 the the DET fcis-4086 42 23 final final ADJ fcis-4086 42 24 layer layer NOUN fcis-4086 42 25 for for ADP fcis-4086 42 26 the the DET fcis-4086 42 27 corresponding correspond VERB fcis-4086 42 28 operations operation NOUN fcis-4086 42 29 . . PUNCT fcis-4086 43 1 in in ADP fcis-4086 43 2 this this DET fcis-4086 43 3 way way NOUN fcis-4086 43 4 , , PUNCT fcis-4086 43 5 deep deep ADJ fcis-4086 43 6 learning learn VERB fcis-4086 43 7 feature feature NOUN fcis-4086 43 8 extractors extractor NOUN fcis-4086 43 9 such such ADJ fcis-4086 43 10 as as ADP fcis-4086 43 11 textcnn textcnn PROPN fcis-4086 43 12 , , PUNCT fcis-4086 43 13 textrnn textrnn NOUN fcis-4086 43 14 , , PUNCT fcis-4086 43 15 etc etc X fcis-4086 43 16 . . X fcis-4086 43 17 can can AUX fcis-4086 43 18 be be AUX fcis-4086 43 19 directly directly ADV fcis-4086 43 20 utilized utilize VERB fcis-4086 43 21 instead instead ADV fcis-4086 43 22 of of ADP fcis-4086 43 23 the the DET fcis-4086 43 24 above above ADJ fcis-4086 43 25 complex complex ADJ fcis-4086 43 26 traditional traditional ADJ fcis-4086 43 27 machine machine NOUN fcis-4086 43 28 learning learn VERB fcis-4086 43 29 feature feature NOUN fcis-4086 43 30 extraction extraction NOUN fcis-4086 43 31 process process NOUN fcis-4086 43 32 . . PUNCT fcis-4086 44 1 textcnn textcnn PROPN fcis-4086 44 2 : : PUNCT fcis-4086 44 3 the the DET fcis-4086 44 4 word word NOUN fcis-4086 44 5 vector vector NOUN fcis-4086 44 6 is be AUX fcis-4086 44 7 first first ADV fcis-4086 44 8 represented represent VERB fcis-4086 44 9 by by ADP fcis-4086 44 10 e(wi e(wi NOUN fcis-4086 44 11 ) ) PUNCT fcis-4086 44 12 . . PUNCT fcis-4086 45 1 then then ADV fcis-4086 45 2 a a DET fcis-4086 45 3 bidirectional bidirectional ADJ fcis-4086 45 4 gru gru PROPN fcis-4086 45 5 is be AUX fcis-4086 45 6 used use VERB fcis-4086 45 7 to to PART fcis-4086 45 8 obtain obtain VERB fcis-4086 45 9 the the DET fcis-4086 45 10 context context NOUN fcis-4086 45 11 vector vector NOUN fcis-4086 45 12 representation representation NOUN fcis-4086 45 13 cl(wi),cr(wi cl(wi),cr(wi NOUN fcis-4086 45 14 ) ) PUNCT fcis-4086 45 15 for for ADP fcis-4086 45 16 each each DET fcis-4086 45 17 word word NOUN fcis-4086 45 18 , , PUNCT fcis-4086 45 19 and and CCONJ fcis-4086 45 20 the the DET fcis-4086 45 21 original original ADJ fcis-4086 45 22 word word NOUN fcis-4086 45 23 vector vector NOUN fcis-4086 45 24 and and CCONJ fcis-4086 45 25 the the DET fcis-4086 45 26 context context NOUN fcis-4086 45 27 representation representation NOUN fcis-4086 45 28 vector vector NOUN fcis-4086 45 29 are be AUX fcis-4086 45 30 aggregated aggregate VERB fcis-4086 45 31 through through ADP fcis-4086 45 32 a a DET fcis-4086 45 33 fully fully ADV fcis-4086 45 34 connected connected ADJ fcis-4086 45 35 network network NOUN fcis-4086 45 36 to to PART fcis-4086 45 37 form form VERB fcis-4086 45 38 a a DET fcis-4086 45 39 new new ADJ fcis-4086 45 40 word word NOUN fcis-4086 45 41 vector vector NOUN fcis-4086 45 42 representation representation NOUN fcis-4086 45 43 . . PUNCT fcis-4086 46 1 after after ADP fcis-4086 46 2 that that PRON fcis-4086 46 3 , , PUNCT fcis-4086 46 4 maximum maximum ADJ fcis-4086 46 5 pooling pooling NOUN fcis-4086 46 6 is be AUX fcis-4086 46 7 used use VERB fcis-4086 46 8 to to PART fcis-4086 46 9 obtain obtain VERB fcis-4086 46 10 the the DET fcis-4086 46 11 final final ADJ fcis-4086 46 12 representation representation NOUN fcis-4086 46 13 of of ADP fcis-4086 46 14 the the DET fcis-4086 46 15 sentences sentence NOUN fcis-4086 46 16 . . PUNCT fcis-4086 47 1 finally finally ADV fcis-4086 47 2 , , PUNCT fcis-4086 47 3 softmax softmax PROPN fcis-4086 47 4 is be AUX fcis-4086 47 5 used use VERB fcis-4086 47 6 to to PART fcis-4086 47 7 do do VERB fcis-4086 47 8 the the DET fcis-4086 47 9 classification classification NOUN fcis-4086 47 10 work work NOUN fcis-4086 47 11 . . PUNCT fcis-4086 48 1 textrnn textrnn NOUN fcis-4086 48 2 : : PUNCT fcis-4086 48 3 firstly firstly ADV fcis-4086 48 4 , , PUNCT fcis-4086 48 5 the the DET fcis-4086 48 6 word word NOUN fcis-4086 48 7 vector vector NOUN fcis-4086 48 8 is be AUX fcis-4086 48 9 represented represent VERB fcis-4086 48 10 by by ADP fcis-4086 48 11 e(wi e(wi NOUN fcis-4086 48 12 ) ) PUNCT fcis-4086 48 13 . . PUNCT fcis-4086 49 1 firstly firstly ADV fcis-4086 49 2 , , PUNCT fcis-4086 49 3 the the DET fcis-4086 49 4 word word NOUN fcis-4086 49 5 vector vector NOUN fcis-4086 49 6 of of ADP fcis-4086 49 7 each each DET fcis-4086 49 8 word word NOUN fcis-4086 49 9 in in ADP fcis-4086 49 10 the the DET fcis-4086 49 11 sentence sentence NOUN fcis-4086 49 12 is be AUX fcis-4086 49 13 input input NOUN fcis-4086 49 14 to to ADP fcis-4086 49 15 the the DET fcis-4086 49 16 bidirectional bidirectional ADJ fcis-4086 49 17 two two NUM fcis-4086 49 18 - - PUNCT fcis-4086 49 19 layer layer NOUN fcis-4086 49 20 lstm lstm NOUN fcis-4086 49 21 in in ADP fcis-4086 49 22 turn turn NOUN fcis-4086 49 23 . . PUNCT fcis-4086 50 1 then then ADV fcis-4086 50 2 the the DET fcis-4086 50 3 hidden hidden ADJ fcis-4086 50 4 layers layer NOUN fcis-4086 50 5 at at ADP fcis-4086 50 6 the the DET fcis-4086 50 7 last last ADJ fcis-4086 50 8 valid valid ADJ fcis-4086 50 9 position position NOUN fcis-4086 50 10 in in ADP fcis-4086 50 11 each each PRON fcis-4086 50 12 of of ADP fcis-4086 50 13 the the DET fcis-4086 50 14 two two NUM fcis-4086 50 15 directions direction NOUN fcis-4086 50 16 are be AUX fcis-4086 50 17 stitched stitch VERB fcis-4086 50 18 into into ADP fcis-4086 50 19 a a DET fcis-4086 50 20 vector vector NOUN fcis-4086 50 21 as as ADP fcis-4086 50 22 a a DET fcis-4086 50 23 representation representation NOUN fcis-4086 50 24 of of ADP fcis-4086 50 25 the the DET fcis-4086 50 26 text text NOUN fcis-4086 50 27 . . PUNCT fcis-4086 51 1 finally finally ADV fcis-4086 51 2 , , PUNCT fcis-4086 51 3 softmax softmax PROPN fcis-4086 51 4 is be AUX fcis-4086 51 5 used use VERB fcis-4086 51 6 to to PART fcis-4086 51 7 do do VERB fcis-4086 51 8 the the DET fcis-4086 51 9 classification classification NOUN fcis-4086 51 10 work work NOUN fcis-4086 51 11 . . PUNCT fcis-4086 52 1 textrcnn textrcnn PROPN fcis-4086 52 2 : : PUNCT fcis-4086 52 3 firstly firstly ADV fcis-4086 52 4 , , PUNCT fcis-4086 52 5 the the DET fcis-4086 52 6 word word NOUN fcis-4086 52 7 vector vector NOUN fcis-4086 52 8 is be AUX fcis-4086 52 9 represented represent VERB fcis-4086 52 10 by by ADP fcis-4086 52 11 e(wi e(wi NOUN fcis-4086 52 12 ) ) PUNCT fcis-4086 52 13 . . PUNCT fcis-4086 53 1 next next ADV fcis-4086 53 2 , , PUNCT fcis-4086 53 3 the the DET fcis-4086 53 4 word word NOUN fcis-4086 53 5 vector vector NOUN fcis-4086 53 6 is be AUX fcis-4086 53 7 passed pass VERB fcis-4086 53 8 through through ADP fcis-4086 53 9 the the DET fcis-4086 53 10 bidirectional bidirectional ADJ fcis-4086 53 11 rnn rnn NOUN fcis-4086 53 12 to to PART fcis-4086 53 13 get get VERB fcis-4086 53 14 cl(wi cl(wi ADJ fcis-4086 53 15 ) ) PUNCT fcis-4086 53 16 and and CCONJ fcis-4086 53 17 cr(wi cr(wi ADJ fcis-4086 53 18 ) ) PUNCT fcis-4086 53 19 vectors vector NOUN fcis-4086 53 20 . . PUNCT fcis-4086 54 1 then then ADV fcis-4086 54 2 the the DET fcis-4086 54 3 cl(wi cl(wi NOUN fcis-4086 54 4 ) ) PUNCT fcis-4086 54 5 , , PUNCT fcis-4086 54 6 e(wi e(wi ADJ fcis-4086 54 7 ) ) PUNCT fcis-4086 54 8 and and CCONJ fcis-4086 54 9 cr(wi cr(wi ADJ fcis-4086 54 10 ) ) PUNCT fcis-4086 54 11 are be AUX fcis-4086 54 12 spliced splice VERB fcis-4086 54 13 to to PART fcis-4086 54 14 obtain obtain VERB fcis-4086 54 15 the the DET fcis-4086 54 16 new new ADJ fcis-4086 54 17 vector vector NOUN fcis-4086 54 18 , , PUNCT fcis-4086 54 19 which which PRON fcis-4086 54 20 is be AUX fcis-4086 54 21 fed feed VERB fcis-4086 54 22 to to ADP fcis-4086 54 23 the the DET fcis-4086 54 24 fully fully ADV fcis-4086 54 25 connected connected ADJ fcis-4086 54 26 network network NOUN fcis-4086 54 27 for for ADP fcis-4086 54 28 integration integration NOUN fcis-4086 54 29 . . PUNCT fcis-4086 55 1 the the DET fcis-4086 55 2 output output NOUN fcis-4086 55 3 of of ADP fcis-4086 55 4 the the DET fcis-4086 55 5 fully fully ADV fcis-4086 55 6 connected connected ADJ fcis-4086 55 7 network network NOUN fcis-4086 55 8 is be AUX fcis-4086 55 9 then then ADV fcis-4086 55 10 subjected subject VERB fcis-4086 55 11 to to ADP fcis-4086 55 12 maxpooling maxpoole VERB fcis-4086 55 13 pooling pool VERB fcis-4086 55 14 operation operation NOUN fcis-4086 55 15 . . PUNCT fcis-4086 56 1 finally finally ADV fcis-4086 56 2 , , PUNCT fcis-4086 56 3 softmax softmax PROPN fcis-4086 56 4 is be AUX fcis-4086 56 5 used use VERB fcis-4086 56 6 to to PART fcis-4086 56 7 do do VERB fcis-4086 56 8 the the DET fcis-4086 56 9 classification classification NOUN fcis-4086 56 10 work work NOUN fcis-4086 56 11 . . PUNCT fcis-4086 57 1 2.3 2.3 NUM fcis-4086 57 2 . . PUNCT fcis-4086 58 1 feature feature NOUN fcis-4086 58 2 representation representation NOUN fcis-4086 58 3 since since SCONJ fcis-4086 58 4 the the DET fcis-4086 58 5 length length NOUN fcis-4086 58 6 of of ADP fcis-4086 58 7 text text NOUN fcis-4086 58 8 is be AUX fcis-4086 58 9 not not PART fcis-4086 58 10 fixed fix VERB fcis-4086 58 11 , , PUNCT fcis-4086 58 12 the the DET fcis-4086 58 13 text text NOUN fcis-4086 58 14 of of ADP fcis-4086 58 15 indefinite indefinite ADJ fcis-4086 58 16 length length NOUN fcis-4086 58 17 should should AUX fcis-4086 58 18 be be AUX fcis-4086 58 19 converted convert VERB fcis-4086 58 20 into into ADP fcis-4086 58 21 a a DET fcis-4086 58 22 fixed fix VERB fcis-4086 58 23 length length NOUN fcis-4086 58 24 space space NOUN fcis-4086 58 25 , , PUNCT fcis-4086 58 26 which which PRON fcis-4086 58 27 needs need VERB fcis-4086 58 28 to to PART fcis-4086 58 29 be be AUX fcis-4086 58 30 represented represent VERB fcis-4086 58 31 after after ADP fcis-4086 58 32 a a DET fcis-4086 58 33 word word NOUN fcis-4086 58 34 embedding embed VERB fcis-4086 58 35 process process NOUN fcis-4086 58 36 . . PUNCT fcis-4086 59 1 if if SCONJ fcis-4086 59 2 the the DET fcis-4086 59 3 words word NOUN fcis-4086 59 4 are be AUX fcis-4086 59 5 represented represent VERB fcis-4086 59 6 as as ADP fcis-4086 59 7 discrete discrete ADJ fcis-4086 59 8 vectors vector NOUN fcis-4086 59 9 by by ADP fcis-4086 59 10 the the DET fcis-4086 59 11 one one NUM fcis-4086 59 12 - - PUNCT fcis-4086 59 13 hot hot ADJ fcis-4086 59 14 method method NOUN fcis-4086 59 15 , , PUNCT fcis-4086 59 16 each each DET fcis-4086 59 17 word word NOUN fcis-4086 59 18 in in ADP fcis-4086 59 19 a a DET fcis-4086 59 20 sentence sentence NOUN fcis-4086 59 21 of of ADP fcis-4086 59 22 n n DET fcis-4086 59 23 words word NOUN fcis-4086 59 24 is be AUX fcis-4086 59 25 converted convert VERB fcis-4086 59 26 into into ADP fcis-4086 59 27 an an DET fcis-4086 59 28 n n ADV fcis-4086 59 29 - - PUNCT fcis-4086 59 30 dimensional dimensional ADJ fcis-4086 59 31 sparse sparse ADJ fcis-4086 59 32 vector vector NOUN fcis-4086 59 33 , , PUNCT fcis-4086 59 34 so so SCONJ fcis-4086 59 35 that that SCONJ fcis-4086 59 36 a a DET fcis-4086 59 37 sentence sentence NOUN fcis-4086 59 38 is be AUX fcis-4086 59 39 converted convert VERB fcis-4086 59 40 into into ADP fcis-4086 59 41 an an DET fcis-4086 59 42 n n NUM fcis-4086 59 43 Γ— Γ— NOUN fcis-4086 59 44 n n CCONJ fcis-4086 59 45 sparse sparse ADJ fcis-4086 59 46 matrix matrix NOUN fcis-4086 59 47 . . PUNCT fcis-4086 60 1 as as SCONJ fcis-4086 60 2 can can AUX fcis-4086 60 3 be be AUX fcis-4086 60 4 seen see VERB fcis-4086 60 5 from from ADP fcis-4086 60 6 the the DET fcis-4086 60 7 above above ADJ fcis-4086 60 8 process process NOUN fcis-4086 60 9 , , PUNCT fcis-4086 60 10 the the DET fcis-4086 60 11 main main ADJ fcis-4086 60 12 problem problem NOUN fcis-4086 60 13 of of ADP fcis-4086 60 14 traditional traditional ADJ fcis-4086 60 15 classification classification NOUN fcis-4086 60 16 is be AUX fcis-4086 60 17 that that SCONJ fcis-4086 60 18 the the DET fcis-4086 60 19 final final ADJ fcis-4086 60 20 text text NOUN fcis-4086 60 21 representation representation NOUN fcis-4086 60 22 obtained obtain VERB fcis-4086 60 23 is be AUX fcis-4086 60 24 highly highly ADV fcis-4086 60 25 sparse sparse ADJ fcis-4086 60 26 and and CCONJ fcis-4086 60 27 high high ADJ fcis-4086 60 28 latitude latitude NOUN fcis-4086 60 29 , , PUNCT fcis-4086 60 30 which which PRON fcis-4086 60 31 makes make VERB fcis-4086 60 32 the the DET fcis-4086 60 33 model model NOUN fcis-4086 60 34 feature feature NOUN fcis-4086 60 35 representation representation NOUN fcis-4086 60 36 weak weak ADJ fcis-4086 60 37 . . PUNCT fcis-4086 61 1 in in ADP fcis-4086 61 2 addition addition NOUN fcis-4086 61 3 , , PUNCT fcis-4086 61 4 feature feature NOUN fcis-4086 61 5 engineering engineering NOUN fcis-4086 61 6 often often ADV fcis-4086 61 7 requires require VERB fcis-4086 61 8 manual manual ADJ fcis-4086 61 9 creation creation NOUN fcis-4086 61 10 , , PUNCT fcis-4086 61 11 and and CCONJ fcis-4086 61 12 the the DET fcis-4086 61 13 whole whole ADJ fcis-4086 61 14 process process NOUN fcis-4086 61 15 is be AUX fcis-4086 61 16 tedious tedious ADJ fcis-4086 61 17 . . PUNCT fcis-4086 62 1 and and CCONJ fcis-4086 62 2 one one NUM fcis-4086 62 3 of of ADP fcis-4086 62 4 the the DET fcis-4086 62 5 important important ADJ fcis-4086 62 6 reasons reason NOUN fcis-4086 62 7 why why SCONJ fcis-4086 62 8 deep deep ADJ fcis-4086 62 9 learning learning NOUN fcis-4086 62 10 has have AUX fcis-4086 62 11 been be AUX fcis-4086 62 12 so so ADV fcis-4086 62 13 successful successful ADJ fcis-4086 62 14 in in ADP fcis-4086 62 15 the the DET fcis-4086 62 16 image image NOUN fcis-4086 62 17 field field NOUN fcis-4086 62 18 is be AUX fcis-4086 62 19 that that SCONJ fcis-4086 62 20 the the DET fcis-4086 62 21 original original ADJ fcis-4086 62 22 data datum NOUN fcis-4086 62 23 of of ADP fcis-4086 62 24 images image NOUN fcis-4086 62 25 are be AUX fcis-4086 62 26 continuous continuous ADJ fcis-4086 62 27 and and CCONJ fcis-4086 62 28 dense dense ADJ fcis-4086 62 29 with with ADP fcis-4086 62 30 local local ADJ fcis-4086 62 31 relevance relevance NOUN fcis-4086 62 32 . . PUNCT fcis-4086 63 1 it it PRON fcis-4086 63 2 is be AUX fcis-4086 63 3 envisioned envision VERB fcis-4086 63 4 that that SCONJ fcis-4086 63 5 if if SCONJ fcis-4086 63 6 deep deep ADJ fcis-4086 63 7 learning learning NOUN fcis-4086 63 8 is be AUX fcis-4086 63 9 applied apply VERB fcis-4086 63 10 to to ADP fcis-4086 63 11 text text NOUN fcis-4086 63 12 classification classification NOUN fcis-4086 63 13 with with ADP fcis-4086 63 14 high high ADJ fcis-4086 63 15 local local ADJ fcis-4086 63 16 relevance relevance NOUN fcis-4086 63 17 the the DET fcis-4086 63 18 problem problem NOUN fcis-4086 63 19 of of ADP fcis-4086 63 20 large large ADJ fcis-4086 63 21 - - PUNCT fcis-4086 63 22 scale scale NOUN fcis-4086 63 23 text text NOUN fcis-4086 63 24 representation representation NOUN fcis-4086 63 25 can can AUX fcis-4086 63 26 be be AUX fcis-4086 63 27 solved solve VERB fcis-4086 63 28 . . PUNCT fcis-4086 64 1 3 3 X fcis-4086 64 2 . . X fcis-4086 64 3 experimental experimental ADJ fcis-4086 64 4 results result NOUN fcis-4086 64 5 test test NOUN fcis-4086 64 6 set set NOUN fcis-4086 64 7 : : PUNCT fcis-4086 64 8 200,000 200,000 NUM fcis-4086 64 9 news news NOUN fcis-4086 64 10 headlines headline NOUN fcis-4086 64 11 from from ADP fcis-4086 64 12 thucnews thucnews NOUN fcis-4086 64 13 with with ADP fcis-4086 64 14 text text NOUN fcis-4086 64 15 lengths length NOUN fcis-4086 64 16 between between ADP fcis-4086 64 17 20 20 NUM fcis-4086 64 18 and and CCONJ fcis-4086 64 19 30 30 NUM fcis-4086 64 20 , , PUNCT fcis-4086 64 21 100,000 100,000 NUM fcis-4086 64 22 categories category NOUN fcis-4086 64 23 in in ADP fcis-4086 64 24 total total ADJ fcis-4086 64 25 , , PUNCT fcis-4086 64 26 20,000 20,000 NUM fcis-4086 64 27 items item NOUN fcis-4086 64 28 per per ADP fcis-4086 64 29 category category NOUN fcis-4086 64 30 . . PUNCT fcis-4086 65 1 operating operate VERB fcis-4086 65 2 system system NOUN fcis-4086 65 3 : : PUNCT fcis-4086 65 4 windows window VERB fcis-4086 65 5 10 10 NUM fcis-4086 65 6 compiler compiler NOUN fcis-4086 65 7 : : PUNCT fcis-4086 65 8 pycharm pycharm NOUN fcis-4086 65 9 community community NOUN fcis-4086 65 10 edition edition NOUN fcis-4086 65 11 2020.2.3 2020.2.3 NUM fcis-4086 65 12 x64 x64 NOUN fcis-4086 65 13 deep deep ADJ fcis-4086 65 14 learning learn VERB fcis-4086 65 15 environment environment NOUN fcis-4086 65 16 : : PUNCT fcis-4086 65 17 pytorch pytorch NOUN fcis-4086 65 18 1.6.0 1.6.0 NUM fcis-4086 65 19 due due ADP fcis-4086 65 20 to to ADP fcis-4086 65 21 the the DET fcis-4086 65 22 large large ADJ fcis-4086 65 23 number number NOUN fcis-4086 65 24 of of ADP fcis-4086 65 25 iterations iteration NOUN fcis-4086 65 26 , , PUNCT fcis-4086 65 27 only only ADV fcis-4086 65 28 the the DET fcis-4086 65 29 first first ADJ fcis-4086 65 30 three three NUM fcis-4086 65 31 iterations iteration NOUN fcis-4086 65 32 are be AUX fcis-4086 65 33 taken take VERB fcis-4086 65 34 here here ADV fcis-4086 65 35 for for ADP fcis-4086 65 36 reference reference NOUN fcis-4086 65 37 , , PUNCT fcis-4086 65 38 and and CCONJ fcis-4086 65 39 the the DET fcis-4086 65 40 status status NOUN fcis-4086 65 41 of of ADP fcis-4086 65 42 each each DET fcis-4086 65 43 model model NOUN fcis-4086 65 44 iteration iteration NOUN fcis-4086 65 45 is be AUX fcis-4086 65 46 as as SCONJ fcis-4086 65 47 follows follow VERB fcis-4086 65 48 : : PUNCT fcis-4086 65 49 figure figure NOUN fcis-4086 65 50 2 2 NUM fcis-4086 65 51 . . PUNCT fcis-4086 65 52 textcnn textcnn PROPN fcis-4086 65 53 loss loss NOUN fcis-4086 65 54 descent descent NOUN fcis-4086 65 55 graph graph NOUN fcis-4086 65 56 : : PUNCT fcis-4086 65 57 blue blue ADJ fcis-4086 65 58 for for ADP fcis-4086 65 59 the the DET fcis-4086 65 60 first first ADJ fcis-4086 65 61 iteration iteration NOUN fcis-4086 65 62 , , PUNCT fcis-4086 65 63 yellow yellow ADJ fcis-4086 65 64 for for ADP fcis-4086 65 65 the the DET fcis-4086 65 66 second second ADJ fcis-4086 65 67 , , PUNCT fcis-4086 65 68 green green ADJ fcis-4086 65 69 for for ADP fcis-4086 65 70 the the DET fcis-4086 65 71 third third ADJ fcis-4086 65 72 60 60 NUM fcis-4086 65 73 figure figure NOUN fcis-4086 65 74 3 3 NUM fcis-4086 65 75 . . PUNCT fcis-4086 65 76 textcnn textcnn PROPN fcis-4086 65 77 accuracy accuracy NOUN fcis-4086 65 78 transformation transformation NOUN fcis-4086 65 79 graph graph NOUN fcis-4086 65 80 : : PUNCT fcis-4086 65 81 blue blue ADJ fcis-4086 65 82 for for ADP fcis-4086 65 83 the the DET fcis-4086 65 84 first first ADJ fcis-4086 65 85 iteration iteration NOUN fcis-4086 65 86 , , PUNCT fcis-4086 65 87 yellow yellow ADJ fcis-4086 65 88 for for ADP fcis-4086 65 89 the the DET fcis-4086 65 90 second second ADJ fcis-4086 65 91 , , PUNCT fcis-4086 65 92 and and CCONJ fcis-4086 65 93 green green ADJ fcis-4086 65 94 for for ADP fcis-4086 65 95 the the DET fcis-4086 65 96 third third ADJ fcis-4086 65 97 the the DET fcis-4086 65 98 following follow VERB fcis-4086 65 99 results result NOUN fcis-4086 65 100 are be AUX fcis-4086 65 101 obtained obtain VERB fcis-4086 65 102 by by ADP fcis-4086 65 103 averaging average VERB fcis-4086 65 104 the the DET fcis-4086 65 105 first first ADJ fcis-4086 65 106 three three NUM fcis-4086 65 107 iterations iteration NOUN fcis-4086 65 108 using use VERB fcis-4086 65 109 the the DET fcis-4086 65 110 textcnn textcnn PROPN fcis-4086 65 111 model model NOUN fcis-4086 65 112 : : PUNCT fcis-4086 65 113 table table NOUN fcis-4086 65 114 1 1 PROPN fcis-4086 65 115 . . PUNCT fcis-4086 66 1 cnn cnn PROPN fcis-4086 66 2 mean mean VERB fcis-4086 66 3 loss loss PROPN fcis-4086 66 4 acc acc PROPN fcis-4086 66 5 first first ADV fcis-4086 66 6 0.586 0.586 NUM fcis-4086 66 7 81.39 81.39 NUM fcis-4086 66 8 % % NOUN fcis-4086 66 9 second second ADJ fcis-4086 66 10 0.3414 0.3414 NUM fcis-4086 66 11 89.47 89.47 NUM fcis-4086 66 12 % % NOUN fcis-4086 66 13 third third NOUN fcis-4086 66 14 0.3307 0.3307 NUM fcis-4086 66 15 89.89 89.89 NUM fcis-4086 66 16 % % NOUN fcis-4086 66 17 figure figure NOUN fcis-4086 66 18 4 4 NUM fcis-4086 66 19 . . PUNCT fcis-4086 66 20 textrnn textrnn PROPN fcis-4086 66 21 accuracy accuracy NOUN fcis-4086 66 22 transformation transformation NOUN fcis-4086 66 23 graph graph NOUN fcis-4086 66 24 : : PUNCT fcis-4086 66 25 blue blue ADJ fcis-4086 66 26 for for ADP fcis-4086 66 27 the the DET fcis-4086 66 28 first first ADJ fcis-4086 66 29 iteration iteration NOUN fcis-4086 66 30 , , PUNCT fcis-4086 66 31 yellow yellow ADJ fcis-4086 66 32 for for ADP fcis-4086 66 33 the the DET fcis-4086 66 34 second second ADJ fcis-4086 66 35 , , PUNCT fcis-4086 66 36 green green ADJ fcis-4086 66 37 for for ADP fcis-4086 66 38 the the DET fcis-4086 66 39 third third ADJ fcis-4086 66 40 figure figure NOUN fcis-4086 66 41 5 5 NUM fcis-4086 66 42 . . PUNCT fcis-4086 67 1 textrnn textrnn PROPN fcis-4086 67 2 accuracy accuracy NOUN fcis-4086 67 3 transformation transformation NOUN fcis-4086 67 4 graph graph NOUN fcis-4086 67 5 : : PUNCT fcis-4086 67 6 blue blue ADJ fcis-4086 67 7 for for ADP fcis-4086 67 8 the the DET fcis-4086 67 9 first first ADJ fcis-4086 67 10 iteration iteration NOUN fcis-4086 67 11 , , PUNCT fcis-4086 67 12 yellow yellow ADJ fcis-4086 67 13 for for ADP fcis-4086 67 14 the the DET fcis-4086 67 15 second second ADJ fcis-4086 67 16 , , PUNCT fcis-4086 67 17 green green ADJ fcis-4086 67 18 for for ADP fcis-4086 67 19 the the DET fcis-4086 67 20 third third ADJ fcis-4086 67 21 the the DET fcis-4086 67 22 following following ADJ fcis-4086 67 23 results result NOUN fcis-4086 67 24 were be AUX fcis-4086 67 25 obtained obtain VERB fcis-4086 67 26 by by ADP fcis-4086 67 27 averaging average VERB fcis-4086 67 28 the the DET fcis-4086 67 29 first first ADJ fcis-4086 67 30 three three NUM fcis-4086 67 31 iterations iteration NOUN fcis-4086 67 32 using use VERB fcis-4086 67 33 the the DET fcis-4086 67 34 textrnn textrnn PROPN fcis-4086 67 35 model model NOUN fcis-4086 67 36 . . PUNCT fcis-4086 68 1 table table NOUN fcis-4086 68 2 2 2 NUM fcis-4086 68 3 . . PUNCT fcis-4086 68 4 rnn rnn VERB fcis-4086 68 5 average average ADJ fcis-4086 68 6 loss loss NOUN fcis-4086 68 7 acc acc PROPN fcis-4086 68 8 first first ADJ fcis-4086 68 9 time time NOUN fcis-4086 68 10 0.7613 0.7613 NUM fcis-4086 68 11 73.55 73.55 NUM fcis-4086 68 12 % % NOUN fcis-4086 68 13 second second ADJ fcis-4086 68 14 time time NOUN fcis-4086 68 15 0.3536 0.3536 NUM fcis-4086 68 16 88.75 88.75 NUM fcis-4086 68 17 % % NOUN fcis-4086 68 18 third third ADJ fcis-4086 68 19 time time NOUN fcis-4086 68 20 0.3229 0.3229 NUM fcis-4086 68 21 89.75 89.75 NUM fcis-4086 68 22 % % NOUN fcis-4086 68 23 figure figure NOUN fcis-4086 68 24 6 6 NUM fcis-4086 68 25 . . PUNCT fcis-4086 69 1 textrcnn textrcnn PROPN fcis-4086 69 2 loss loss NOUN fcis-4086 69 3 descent descent NOUN fcis-4086 69 4 graph graph NOUN fcis-4086 69 5 : : PUNCT fcis-4086 69 6 blue blue ADJ fcis-4086 69 7 for for ADP fcis-4086 69 8 the the DET fcis-4086 69 9 first first ADJ fcis-4086 69 10 iteration iteration NOUN fcis-4086 69 11 , , PUNCT fcis-4086 69 12 yellow yellow ADJ fcis-4086 69 13 for for ADP fcis-4086 69 14 the the DET fcis-4086 69 15 second second ADJ fcis-4086 69 16 , , PUNCT fcis-4086 69 17 green green ADJ fcis-4086 69 18 for for ADP fcis-4086 69 19 the the DET fcis-4086 69 20 third third ADJ fcis-4086 69 21 figure figure NOUN fcis-4086 69 22 7 7 NUM fcis-4086 69 23 . . PUNCT fcis-4086 70 1 textrcnn textrcnn PROPN fcis-4086 70 2 accuracy accuracy NOUN fcis-4086 70 3 transformation transformation NOUN fcis-4086 70 4 graph graph NOUN fcis-4086 70 5 : : PUNCT fcis-4086 70 6 blue blue ADJ fcis-4086 70 7 for for ADP fcis-4086 70 8 the the DET fcis-4086 70 9 first first ADJ fcis-4086 70 10 iteration iteration NOUN fcis-4086 70 11 , , PUNCT fcis-4086 70 12 yellow yellow ADJ fcis-4086 70 13 for for ADP fcis-4086 70 14 the the DET fcis-4086 70 15 second second ADJ fcis-4086 70 16 , , PUNCT fcis-4086 70 17 green green ADJ fcis-4086 70 18 for for ADP fcis-4086 70 19 the the DET fcis-4086 70 20 third third ADJ fcis-4086 70 21 the the DET fcis-4086 70 22 average average NOUN fcis-4086 70 23 of of ADP fcis-4086 70 24 the the DET fcis-4086 70 25 first first ADJ fcis-4086 70 26 three three NUM fcis-4086 70 27 iterations iteration NOUN fcis-4086 70 28 using use VERB fcis-4086 70 29 the the DET fcis-4086 70 30 textrcnn textrcnn ADJ fcis-4086 70 31 model model NOUN fcis-4086 70 32 gives give VERB fcis-4086 70 33 the the DET fcis-4086 70 34 following follow VERB fcis-4086 70 35 results result NOUN fcis-4086 70 36 : : PUNCT fcis-4086 70 37 table table NOUN fcis-4086 70 38 3 3 PROPN fcis-4086 70 39 . . PUNCT fcis-4086 70 40 rcnn rcnn PROPN fcis-4086 70 41 average average PROPN fcis-4086 70 42 loss loss PROPN fcis-4086 70 43 acc acc PROPN fcis-4086 70 44 first first ADV fcis-4086 70 45 0.5507 0.5507 NUM fcis-4086 70 46 81.72 81.72 NUM fcis-4086 70 47 % % NOUN fcis-4086 70 48 second second ADJ fcis-4086 70 49 0.3143 0.3143 NUM fcis-4086 70 50 89.94 89.94 NUM fcis-4086 70 51 % % NOUN fcis-4086 70 52 third third NOUN fcis-4086 70 53 0.2979 0.2979 NUM fcis-4086 70 54 90.59 90.59 NUM fcis-4086 70 55 % % NOUN fcis-4086 70 56 a a DET fcis-4086 70 57 confusion confusion NOUN fcis-4086 70 58 matrix matrix NOUN fcis-4086 70 59 is be AUX fcis-4086 70 60 a a DET fcis-4086 70 61 situation situation NOUN fcis-4086 70 62 analysis analysis NOUN fcis-4086 70 63 table table NOUN fcis-4086 70 64 that that PRON fcis-4086 70 65 summarizes summarize VERB fcis-4086 70 66 the the DET fcis-4086 70 67 prediction prediction NOUN fcis-4086 70 68 results result NOUN fcis-4086 70 69 of of ADP fcis-4086 70 70 classification classification NOUN fcis-4086 70 71 models model NOUN fcis-4086 70 72 in in ADP fcis-4086 70 73 data datum NOUN fcis-4086 70 74 science science NOUN fcis-4086 70 75 , , PUNCT fcis-4086 70 76 data data VERB fcis-4086 70 77 analysis analysis NOUN fcis-4086 70 78 and and CCONJ fcis-4086 70 79 machine machine NOUN fcis-4086 70 80 learning learning NOUN fcis-4086 70 81 . . PUNCT fcis-4086 71 1 it it PRON fcis-4086 71 2 summarizes summarize VERB fcis-4086 71 3 the the DET fcis-4086 71 4 records record NOUN fcis-4086 71 5 in in ADP fcis-4086 71 6 a a DET fcis-4086 71 7 data datum NOUN fcis-4086 71 8 set set VERB fcis-4086 71 9 in in ADP fcis-4086 71 10 matrix matrix NOUN fcis-4086 71 11 form form NOUN fcis-4086 71 12 according accord VERB fcis-4086 71 13 to to ADP fcis-4086 71 14 two two NUM fcis-4086 71 15 criteria criterion NOUN fcis-4086 71 16 : : PUNCT fcis-4086 71 17 the the DET fcis-4086 71 18 true true ADJ fcis-4086 71 19 categories category NOUN fcis-4086 71 20 and and CCONJ fcis-4086 71 21 the the DET fcis-4086 71 22 classification classification NOUN fcis-4086 71 23 judgments judgment NOUN fcis-4086 71 24 made make VERB fcis-4086 71 25 by by ADP fcis-4086 71 26 the the DET fcis-4086 71 27 classification classification NOUN fcis-4086 71 28 models model NOUN fcis-4086 71 29 . . PUNCT fcis-4086 72 1 the the DET fcis-4086 72 2 confusion confusion NOUN fcis-4086 72 3 matrix matrix NOUN fcis-4086 72 4 for for ADP fcis-4086 72 5 each each DET fcis-4086 72 6 model model NOUN fcis-4086 72 7 is be AUX fcis-4086 72 8 as as SCONJ fcis-4086 72 9 follows follow VERB fcis-4086 72 10 . . PUNCT fcis-4086 73 1 figure figure VERB fcis-4086 73 2 8 8 NUM fcis-4086 73 3 . . PUNCT fcis-4086 74 1 confusion confusion NOUN fcis-4086 74 2 matrix matrix NOUN fcis-4086 74 3 of of ADP fcis-4086 74 4 textcnn textcnn PROPN fcis-4086 74 5 61 61 NUM fcis-4086 74 6 figure figure NOUN fcis-4086 74 7 9 9 NUM fcis-4086 74 8 . . PUNCT fcis-4086 74 9 confusion confusion NOUN fcis-4086 74 10 matrix matrix NOUN fcis-4086 74 11 of of ADP fcis-4086 74 12 textrnn textrnn NOUN fcis-4086 74 13 figure figure VERB fcis-4086 74 14 10 10 NUM fcis-4086 74 15 . . PUNCT fcis-4086 75 1 confusion confusion NOUN fcis-4086 75 2 matrix matrix NOUN fcis-4086 75 3 of of ADP fcis-4086 75 4 textrnn textrnn NOUN fcis-4086 75 5 the the DET fcis-4086 75 6 confusion confusion NOUN fcis-4086 75 7 matrix matrix NOUN fcis-4086 75 8 shows show VERB fcis-4086 75 9 that that SCONJ fcis-4086 75 10 the the DET fcis-4086 75 11 percentage percentage NOUN fcis-4086 75 12 of of ADP fcis-4086 75 13 correct correct ADJ fcis-4086 75 14 predictions prediction NOUN fcis-4086 75 15 is be AUX fcis-4086 75 16 higher high ADJ fcis-4086 75 17 in in ADP fcis-4086 75 18 the the DET fcis-4086 75 19 total total ADJ fcis-4086 75 20 sample sample NOUN fcis-4086 75 21 , , PUNCT fcis-4086 75 22 with with ADP fcis-4086 75 23 the the DET fcis-4086 75 24 textrcnn textrcnn ADJ fcis-4086 75 25 model model NOUN fcis-4086 75 26 performing perform VERB fcis-4086 75 27 slightly slightly ADV fcis-4086 75 28 better well ADJ fcis-4086 75 29 . . PUNCT fcis-4086 76 1 cross cross NOUN fcis-4086 76 2 entropy entropy PROPN fcis-4086 76 3 loss loss NOUN fcis-4086 76 4 is be AUX fcis-4086 76 5 often often ADV fcis-4086 76 6 used use VERB fcis-4086 76 7 when when SCONJ fcis-4086 76 8 solving solve VERB fcis-4086 76 9 classification classification NOUN fcis-4086 76 10 problems problem NOUN fcis-4086 76 11 using use VERB fcis-4086 76 12 neural neural ADJ fcis-4086 76 13 networks network NOUN fcis-4086 76 14 , , PUNCT fcis-4086 76 15 and and CCONJ fcis-4086 76 16 it it PRON fcis-4086 76 17 is be AUX fcis-4086 76 18 often often ADV fcis-4086 76 19 used use VERB fcis-4086 76 20 to to PART fcis-4086 76 21 determine determine VERB fcis-4086 76 22 how how SCONJ fcis-4086 76 23 close close ADJ fcis-4086 76 24 the the DET fcis-4086 76 25 actual actual ADJ fcis-4086 76 26 output output NOUN fcis-4086 76 27 is be AUX fcis-4086 76 28 to to ADP fcis-4086 76 29 the the DET fcis-4086 76 30 desired desire VERB fcis-4086 76 31 output output NOUN fcis-4086 76 32 . . PUNCT fcis-4086 77 1 the the DET fcis-4086 77 2 formula formula NOUN fcis-4086 77 3 for for ADP fcis-4086 77 4 implementing implement VERB fcis-4086 77 5 cross cross NOUN fcis-4086 77 6 entropy entropy NOUN fcis-4086 77 7 loss loss NOUN fcis-4086 77 8 in in ADP fcis-4086 77 9 pytorch pytorch NOUN fcis-4086 77 10 is be AUX fcis-4086 77 11 : : PUNCT fcis-4086 77 12 πΏπ‘œπ‘”π‘†π‘œπ‘“π‘‘π‘šπ‘Žπ‘₯(π‘₯𝑖 πΏπ‘œπ‘”π‘†π‘œπ‘“π‘‘π‘šπ‘Žπ‘₯(π‘₯𝑖 X fcis-4086 77 13 ) ) PUNCT fcis-4086 77 14 = = PUNCT fcis-4086 77 15 log log NOUN fcis-4086 77 16 exp exp NOUN fcis-4086 77 17 ( ( PUNCT fcis-4086 77 18 π‘₯𝑖 π‘₯𝑖 PROPN fcis-4086 77 19 ) ) PUNCT fcis-4086 77 20 βˆ‘ βˆ‘ NOUN fcis-4086 77 21 exp exp X fcis-4086 77 22 ( ( PUNCT fcis-4086 77 23 π‘₯𝑗)𝑗 π‘₯𝑗)𝑗 PROPN fcis-4086 77 24 the the DET fcis-4086 77 25 accuracy accuracy NOUN fcis-4086 77 26 rate rate NOUN fcis-4086 77 27 is be AUX fcis-4086 77 28 calculated calculate VERB fcis-4086 77 29 as as ADP fcis-4086 77 30 : : PUNCT fcis-4086 77 31 𝐴𝑐𝑐 𝐴𝑐𝑐 PROPN fcis-4086 77 32 = = SYM fcis-4086 77 33 𝑇𝑃 𝑇𝑃 PROPN fcis-4086 77 34 + + CCONJ fcis-4086 77 35 𝑇𝑁 𝑇𝑁 PROPN fcis-4086 77 36 𝑇𝑃 𝑇𝑃 PROPN fcis-4086 77 37 + + CCONJ fcis-4086 77 38 𝐹𝑃 𝐹𝑃 PROPN fcis-4086 77 39 + + CCONJ fcis-4086 78 1 𝑇𝑁 𝑇𝑁 PROPN fcis-4086 78 2 + + CCONJ fcis-4086 78 3 𝐹𝑁 𝐹𝑁 NOUN fcis-4086 78 4 analyzing analyze VERB fcis-4086 78 5 the the DET fcis-4086 78 6 results result NOUN fcis-4086 78 7 of of ADP fcis-4086 78 8 the the DET fcis-4086 78 9 first first ADJ fcis-4086 78 10 three three NUM fcis-4086 78 11 iterations iteration NOUN fcis-4086 78 12 of of ADP fcis-4086 78 13 each each PRON fcis-4086 78 14 of of ADP fcis-4086 78 15 the the DET fcis-4086 78 16 three three NUM fcis-4086 78 17 models model NOUN fcis-4086 78 18 , , PUNCT fcis-4086 78 19 we we PRON fcis-4086 78 20 can can AUX fcis-4086 78 21 see see VERB fcis-4086 78 22 that that SCONJ fcis-4086 78 23 the the DET fcis-4086 78 24 loss loss NOUN fcis-4086 78 25 and and CCONJ fcis-4086 78 26 acc acc PROPN fcis-4086 78 27 values value NOUN fcis-4086 78 28 ( ( PUNCT fcis-4086 78 29 accuracy accuracy NOUN fcis-4086 78 30 refers refer VERB fcis-4086 78 31 to to ADP fcis-4086 78 32 the the DET fcis-4086 78 33 percentage percentage NOUN fcis-4086 78 34 of of ADP fcis-4086 78 35 correct correct ADJ fcis-4086 78 36 predictions prediction NOUN fcis-4086 78 37 over over ADP fcis-4086 78 38 the the DET fcis-4086 78 39 total total ADJ fcis-4086 78 40 sample sample NOUN fcis-4086 78 41 ) ) PUNCT fcis-4086 78 42 are be AUX fcis-4086 78 43 not not PART fcis-4086 78 44 very very ADV fcis-4086 78 45 good good ADJ fcis-4086 78 46 after after ADP fcis-4086 78 47 the the DET fcis-4086 78 48 first first ADJ fcis-4086 78 49 iteration iteration NOUN fcis-4086 78 50 , , PUNCT fcis-4086 78 51 but but CCONJ fcis-4086 78 52 all all DET fcis-4086 78 53 models model NOUN fcis-4086 78 54 have have VERB fcis-4086 78 55 a a DET fcis-4086 78 56 significant significant ADJ fcis-4086 78 57 improvement improvement NOUN fcis-4086 78 58 after after ADP fcis-4086 78 59 the the DET fcis-4086 78 60 second second ADJ fcis-4086 78 61 iteration iteration NOUN fcis-4086 78 62 , , PUNCT fcis-4086 78 63 with with ADP fcis-4086 78 64 lower low ADJ fcis-4086 78 65 loss loss NOUN fcis-4086 78 66 function function NOUN fcis-4086 78 67 values value NOUN fcis-4086 78 68 than than ADP fcis-4086 78 69 the the DET fcis-4086 78 70 first first ADJ fcis-4086 78 71 one one NUM fcis-4086 78 72 and and CCONJ fcis-4086 78 73 a a DET fcis-4086 78 74 corresponding corresponding ADJ fcis-4086 78 75 increase increase NOUN fcis-4086 78 76 in in ADP fcis-4086 78 77 accuracy accuracy NOUN fcis-4086 78 78 . . PUNCT fcis-4086 79 1 the the DET fcis-4086 79 2 difference difference NOUN fcis-4086 79 3 between between ADP fcis-4086 79 4 the the DET fcis-4086 79 5 second second ADJ fcis-4086 79 6 and and CCONJ fcis-4086 79 7 third third ADJ fcis-4086 79 8 iterations iteration NOUN fcis-4086 79 9 was be AUX fcis-4086 79 10 not not PART fcis-4086 79 11 significant significant ADJ fcis-4086 79 12 , , PUNCT fcis-4086 79 13 but but CCONJ fcis-4086 79 14 the the DET fcis-4086 79 15 improvement improvement NOUN fcis-4086 79 16 was be AUX fcis-4086 79 17 generally generally ADV fcis-4086 79 18 steady steady ADJ fcis-4086 79 19 . . PUNCT fcis-4086 80 1 finally finally ADV fcis-4086 80 2 , , PUNCT fcis-4086 80 3 after after ADP fcis-4086 80 4 completing complete VERB fcis-4086 80 5 all all DET fcis-4086 80 6 iterations iteration NOUN fcis-4086 80 7 , , PUNCT fcis-4086 80 8 the the DET fcis-4086 80 9 results result NOUN fcis-4086 80 10 of of ADP fcis-4086 80 11 the the DET fcis-4086 80 12 three three NUM fcis-4086 80 13 models model NOUN fcis-4086 80 14 textcnn textcnn NOUN fcis-4086 80 15 , , PUNCT fcis-4086 80 16 textrnn textrnn NOUN fcis-4086 80 17 , , PUNCT fcis-4086 80 18 and and CCONJ fcis-4086 80 19 textrcnn textrcnn PROPN fcis-4086 80 20 were be AUX fcis-4086 80 21 obtained obtain VERB fcis-4086 80 22 with with ADP fcis-4086 80 23 loss loss NOUN fcis-4086 80 24 values value NOUN fcis-4086 80 25 of of ADP fcis-4086 80 26 0.3 0.3 NUM fcis-4086 80 27 , , PUNCT fcis-4086 80 28 0.29 0.29 NUM fcis-4086 80 29 , , PUNCT fcis-4086 80 30 and and CCONJ fcis-4086 80 31 0.28 0.28 NUM fcis-4086 80 32 , , PUNCT fcis-4086 80 33 respectively respectively ADV fcis-4086 80 34 , , PUNCT fcis-4086 80 35 and and CCONJ fcis-4086 80 36 accuracy accuracy NOUN fcis-4086 80 37 rates rate NOUN fcis-4086 80 38 of of ADP fcis-4086 80 39 90.42 90.42 NUM fcis-4086 80 40 % % NOUN fcis-4086 80 41 , , PUNCT fcis-4086 80 42 90.75 90.75 NUM fcis-4086 80 43 % % NOUN fcis-4086 80 44 , , PUNCT fcis-4086 80 45 and and CCONJ fcis-4086 80 46 91.10 91.10 NUM fcis-4086 80 47 % % NOUN fcis-4086 80 48 , , PUNCT fcis-4086 80 49 respectively respectively ADV fcis-4086 80 50 . . PUNCT fcis-4086 81 1 the the DET fcis-4086 81 2 experimental experimental ADJ fcis-4086 81 3 results result NOUN fcis-4086 81 4 are be AUX fcis-4086 81 5 then then ADV fcis-4086 81 6 compared compare VERB fcis-4086 81 7 to to PART fcis-4086 81 8 obtain obtain VERB fcis-4086 81 9 the the DET fcis-4086 81 10 following follow VERB fcis-4086 81 11 results result NOUN fcis-4086 81 12 : : PUNCT fcis-4086 81 13 figure figure VERB fcis-4086 81 14 11 11 NUM fcis-4086 81 15 . . PUNCT fcis-4086 82 1 three three NUM fcis-4086 82 2 models model NOUN fcis-4086 82 3 loss loss NOUN fcis-4086 82 4 values value NOUN fcis-4086 82 5 figure figure VERB fcis-4086 82 6 12 12 NUM fcis-4086 82 7 . . PUNCT fcis-4086 83 1 final final ADJ fcis-4086 83 2 prediction prediction NOUN fcis-4086 83 3 accuracy accuracy NOUN fcis-4086 83 4 of of ADP fcis-4086 83 5 the the DET fcis-4086 83 6 three three NUM fcis-4086 83 7 models model NOUN fcis-4086 83 8 , , PUNCT fcis-4086 83 9 left leave VERB fcis-4086 83 10 to to ADP fcis-4086 83 11 right right NOUN fcis-4086 83 12 for for ADP fcis-4086 83 13 textcnn textcnn PROPN fcis-4086 83 14 , , PUNCT fcis-4086 83 15 textrnn textrnn NOUN fcis-4086 83 16 , , PUNCT fcis-4086 83 17 textrcnn textrcnn NOUN fcis-4086 83 18 4 4 NUM fcis-4086 83 19 . . PUNCT fcis-4086 84 1 summary summary NOUN fcis-4086 84 2 and and CCONJ fcis-4086 84 3 outlook outlook VERB fcis-4086 84 4 through through ADP fcis-4086 84 5 the the DET fcis-4086 84 6 above above ADJ fcis-4086 84 7 confusion confusion NOUN fcis-4086 84 8 matrix matrix NOUN fcis-4086 84 9 and and CCONJ fcis-4086 84 10 experimental experimental ADJ fcis-4086 84 11 charts chart NOUN fcis-4086 84 12 , , PUNCT fcis-4086 84 13 the the DET fcis-4086 84 14 loss loss NOUN fcis-4086 84 15 value value NOUN fcis-4086 84 16 is be AUX fcis-4086 84 17 lower low ADJ fcis-4086 84 18 than than ADP fcis-4086 84 19 cnn cnn PROPN fcis-4086 84 20 and and CCONJ fcis-4086 84 21 rnn rnn VERB fcis-4086 84 22 0.02 0.02 NUM fcis-4086 84 23 , , PUNCT fcis-4086 84 24 0.01 0.01 NUM fcis-4086 84 25 , , PUNCT fcis-4086 84 26 and and CCONJ fcis-4086 84 27 the the DET fcis-4086 84 28 accuracy accuracy NOUN fcis-4086 84 29 rate rate NOUN fcis-4086 84 30 is be AUX fcis-4086 84 31 higher high ADJ fcis-4086 84 32 than than ADP fcis-4086 84 33 cnn cnn PROPN fcis-4086 84 34 and and CCONJ fcis-4086 84 35 rnn rnn VERB fcis-4086 84 36 0.68 0.68 NUM fcis-4086 84 37 % % NOUN fcis-4086 84 38 , , PUNCT fcis-4086 84 39 0.35 0.35 NUM fcis-4086 84 40 % % NOUN fcis-4086 84 41 , , PUNCT fcis-4086 84 42 indicating indicate VERB fcis-4086 84 43 that that SCONJ fcis-4086 84 44 the the DET fcis-4086 84 45 textrcnn textrcnn PROPN fcis-4086 84 46 model model NOUN fcis-4086 84 47 performs perform VERB fcis-4086 84 48 better well ADJ fcis-4086 84 49 than than ADP fcis-4086 84 50 rnn rnn NOUN fcis-4086 84 51 and and CCONJ fcis-4086 84 52 cnn cnn PROPN fcis-4086 84 53 , , PUNCT fcis-4086 84 54 is be AUX fcis-4086 84 55 easier easy ADJ fcis-4086 84 56 to to PART fcis-4086 84 57 train train VERB fcis-4086 84 58 , , PUNCT fcis-4086 84 59 and and CCONJ fcis-4086 84 60 the the DET fcis-4086 84 61 applicable applicable ADJ fcis-4086 84 62 range range NOUN fcis-4086 84 63 should should AUX fcis-4086 84 64 be be AUX fcis-4086 84 65 broader broad ADJ fcis-4086 84 66 . . PUNCT fcis-4086 85 1 however however ADV fcis-4086 85 2 , , PUNCT fcis-4086 85 3 the the DET fcis-4086 85 4 model model NOUN fcis-4086 85 5 should should AUX fcis-4086 85 6 not not PART fcis-4086 85 7 be be AUX fcis-4086 85 8 selected select VERB fcis-4086 85 9 blindly blindly ADV fcis-4086 85 10 in in ADP fcis-4086 85 11 the the DET fcis-4086 85 12 actual actual ADJ fcis-4086 85 13 training training NOUN fcis-4086 85 14 , , PUNCT fcis-4086 85 15 but but CCONJ fcis-4086 85 16 should should AUX fcis-4086 85 17 be be AUX fcis-4086 85 18 analyzed analyze VERB fcis-4086 85 19 for for ADP fcis-4086 85 20 specific specific ADJ fcis-4086 85 21 problems problem NOUN fcis-4086 85 22 before before ADP fcis-4086 85 23 choosing choose VERB fcis-4086 85 24 a a DET fcis-4086 85 25 suitable suitable ADJ fcis-4086 85 26 model model NOUN fcis-4086 85 27 . . PUNCT fcis-4086 86 1 traditional traditional ADJ fcis-4086 86 2 machine machine NOUN fcis-4086 86 3 learning learning NOUN fcis-4086 86 4 requires require VERB fcis-4086 86 5 a a DET fcis-4086 86 6 lot lot NOUN fcis-4086 86 7 of of ADP fcis-4086 86 8 preprocessing preprocesse VERB fcis-4086 86 9 work work NOUN fcis-4086 86 10 in in ADP fcis-4086 86 11 text text NOUN fcis-4086 86 12 classification classification NOUN fcis-4086 86 13 , , PUNCT fcis-4086 86 14 and and CCONJ fcis-4086 86 15 when when SCONJ fcis-4086 86 16 text text NOUN fcis-4086 86 17 features feature NOUN fcis-4086 86 18 are be AUX fcis-4086 86 19 extracted extract VERB fcis-4086 86 20 , , PUNCT fcis-4086 86 21 more more ADV fcis-4086 86 22 shallow shallow ADJ fcis-4086 86 23 features feature NOUN fcis-4086 86 24 are be AUX fcis-4086 86 25 obtained obtain VERB fcis-4086 86 26 . . PUNCT fcis-4086 87 1 deep deep ADJ fcis-4086 87 2 learning learning NOUN fcis-4086 87 3 models model NOUN fcis-4086 87 4 are be AUX fcis-4086 87 5 generally generally ADV fcis-4086 87 6 better well ADJ fcis-4086 87 7 than than ADP fcis-4086 87 8 machine machine NOUN fcis-4086 87 9 learning learning NOUN fcis-4086 87 10 for for ADP fcis-4086 87 11 text text NOUN fcis-4086 87 12 classification classification NOUN fcis-4086 87 13 , , PUNCT fcis-4086 87 14 but but CCONJ fcis-4086 87 15 there there PRON fcis-4086 87 16 are be VERB fcis-4086 87 17 differences difference NOUN fcis-4086 87 18 between between ADP fcis-4086 87 19 different different ADJ fcis-4086 87 20 models model NOUN fcis-4086 87 21 and and CCONJ fcis-4086 87 22 they they PRON fcis-4086 87 23 all all PRON fcis-4086 87 24 have have VERB fcis-4086 87 25 their their PRON fcis-4086 87 26 own own ADJ fcis-4086 87 27 specialties specialty NOUN fcis-4086 87 28 , , PUNCT fcis-4086 87 29 and and CCONJ fcis-4086 87 30 the the DET fcis-4086 87 31 above above ADJ fcis-4086 87 32 experiments experiment NOUN fcis-4086 87 33 are be AUX fcis-4086 87 34 also also ADV fcis-4086 87 35 concluded conclude VERB fcis-4086 87 36 by by ADP fcis-4086 87 37 comparison comparison NOUN fcis-4086 87 38 . . PUNCT fcis-4086 88 1 in in ADP fcis-4086 88 2 the the DET fcis-4086 88 3 real real ADJ fcis-4086 88 4 problem problem NOUN fcis-4086 88 5 , , PUNCT fcis-4086 88 6 how how SCONJ fcis-4086 88 7 to to PART fcis-4086 88 8 choose choose VERB fcis-4086 88 9 the the DET fcis-4086 88 10 right right ADJ fcis-4086 88 11 model model NOUN fcis-4086 88 12 is be AUX fcis-4086 88 13 also also ADV fcis-4086 88 14 an an DET fcis-4086 88 15 issue issue NOUN fcis-4086 88 16 that that PRON fcis-4086 88 17 needs need VERB fcis-4086 88 18 attention attention NOUN fcis-4086 88 19 . . PUNCT fcis-4086 89 1 the the DET fcis-4086 89 2 significant significant ADJ fcis-4086 89 3 feature feature NOUN fcis-4086 89 4 of of ADP fcis-4086 89 5 deep deep ADJ fcis-4086 89 6 learning learning NOUN fcis-4086 89 7 model model NOUN fcis-4086 89 8 is be AUX fcis-4086 89 9 to to PART fcis-4086 89 10 extract extract VERB fcis-4086 89 11 more more ADJ fcis-4086 89 12 deep deep ADJ fcis-4086 89 13 text text NOUN fcis-4086 89 14 feature feature NOUN fcis-4086 89 15 information information NOUN fcis-4086 89 16 , , PUNCT fcis-4086 89 17 reduce reduce VERB fcis-4086 89 18 the the DET fcis-4086 89 19 loss loss NOUN fcis-4086 89 20 of of ADP fcis-4086 89 21 document document NOUN fcis-4086 89 22 information information NOUN fcis-4086 89 23 and and CCONJ fcis-4086 89 24 improve improve VERB fcis-4086 89 25 the the DET fcis-4086 89 26 accuracy accuracy NOUN fcis-4086 89 27 of of ADP fcis-4086 89 28 text text NOUN fcis-4086 89 29 classification classification NOUN fcis-4086 89 30 , , PUNCT fcis-4086 89 31 but but CCONJ fcis-4086 89 32 gold gold NOUN fcis-4086 89 33 is be AUX fcis-4086 89 34 not not PART fcis-4086 89 35 perfect perfect ADJ fcis-4086 89 36 , , PUNCT fcis-4086 89 37 deep deep ADJ fcis-4086 89 38 learning learning NOUN fcis-4086 89 39 model model NOUN fcis-4086 89 40 is be AUX fcis-4086 89 41 far far ADV fcis-4086 89 42 from from ADP fcis-4086 89 43 perfect perfect ADJ fcis-4086 89 44 , , PUNCT fcis-4086 89 45 the the DET fcis-4086 89 46 main main ADJ fcis-4086 89 47 problem problem NOUN fcis-4086 89 48 is be AUX fcis-4086 89 49 still still ADV fcis-4086 89 50 the the DET fcis-4086 89 51 need need NOUN fcis-4086 89 52 for for ADP fcis-4086 89 53 massive massive ADJ fcis-4086 89 54 data datum NOUN fcis-4086 89 55 scale scale NOUN fcis-4086 89 56 , , PUNCT fcis-4086 89 57 and and CCONJ fcis-4086 89 58 how how SCONJ fcis-4086 89 59 to to PART fcis-4086 89 60 design design VERB fcis-4086 89 61 a a DET fcis-4086 89 62 new new ADJ fcis-4086 89 63 deep deep ADJ fcis-4086 89 64 neural neural ADJ fcis-4086 89 65 network network NOUN fcis-4086 89 66 that that PRON fcis-4086 89 67 can can AUX fcis-4086 89 68 complement complement VERB fcis-4086 89 69 the the DET fcis-4086 89 70 advantages advantage NOUN fcis-4086 89 71 of of ADP fcis-4086 89 72 the the DET fcis-4086 89 73 model model NOUN fcis-4086 89 74 after after ADP fcis-4086 89 75 getting get VERB fcis-4086 89 76 considerable considerable ADJ fcis-4086 89 77 data datum NOUN fcis-4086 89 78 may may AUX fcis-4086 89 79 be be AUX fcis-4086 89 80 a a DET fcis-4086 89 81 research research NOUN fcis-4086 89 82 direction direction NOUN fcis-4086 89 83 that that PRON fcis-4086 89 84 needs need VERB fcis-4086 89 85 to to PART fcis-4086 89 86 be be AUX fcis-4086 89 87 paid pay VERB fcis-4086 89 88 attention attention NOUN fcis-4086 89 89 to to ADP fcis-4086 89 90 . . PUNCT fcis-4086 90 1 there there PRON fcis-4086 90 2 is be VERB fcis-4086 90 3 no no DET fcis-4086 90 4 doubt doubt NOUN fcis-4086 90 5 that that SCONJ fcis-4086 90 6 the the DET fcis-4086 90 7 internet internet NOUN fcis-4086 90 8 will will AUX fcis-4086 90 9 continue continue VERB fcis-4086 90 10 to to PART fcis-4086 90 11 flourish flourish VERB fcis-4086 90 12 in in ADP fcis-4086 90 13 the the DET fcis-4086 90 14 future future NOUN fcis-4086 90 15 , , PUNCT fcis-4086 90 16 and and CCONJ fcis-4086 90 17 this this PRON fcis-4086 90 18 is be AUX fcis-4086 90 19 the the DET fcis-4086 90 20 driving drive VERB fcis-4086 90 21 force force NOUN fcis-4086 90 22 behind behind ADP fcis-4086 90 23 the the DET fcis-4086 90 24 development development NOUN fcis-4086 90 25 of of ADP fcis-4086 90 26 new new ADJ fcis-4086 90 27 technologies technology NOUN fcis-4086 90 28 , , PUNCT fcis-4086 90 29 so so SCONJ fcis-4086 90 30 it it PRON fcis-4086 90 31 can can AUX fcis-4086 90 32 be be AUX fcis-4086 90 33 expected expect VERB fcis-4086 90 34 that that SCONJ fcis-4086 90 35 more more ADV fcis-4086 90 36 perfect perfect ADJ fcis-4086 90 37 new new ADJ fcis-4086 90 38 technologies technology NOUN fcis-4086 90 39 will will AUX fcis-4086 90 40 be be AUX fcis-4086 90 41 applied apply VERB fcis-4086 90 42 to to ADP fcis-4086 90 43 the the DET fcis-4086 90 44 field field NOUN fcis-4086 90 45 of of ADP fcis-4086 90 46 artificial artificial ADJ fcis-4086 90 47 intelligence intelligence NOUN fcis-4086 90 48 in in ADP fcis-4086 90 49 the the DET fcis-4086 90 50 future future NOUN fcis-4086 90 51 . . PUNCT fcis-4086 91 1 62 62 NUM fcis-4086 91 2 references reference NOUN fcis-4086 91 3 [ [ X fcis-4086 91 4 1 1 NUM fcis-4086 91 5 ] ] X fcis-4086 91 6 minaee minaee X fcis-4086 91 7 s s NOUN fcis-4086 91 8 , , PUNCT fcis-4086 91 9 kalchbrenner kalchbrenner NOUN fcis-4086 91 10 n n NOUN fcis-4086 91 11 , , PUNCT fcis-4086 91 12 cambria cambria PROPN fcis-4086 91 13 e e PROPN fcis-4086 91 14 , , PUNCT fcis-4086 91 15 et et PROPN fcis-4086 91 16 al al PROPN fcis-4086 91 17 . . PUNCT fcis-4086 92 1 deep deep ADJ fcis-4086 92 2 learning learning NOUN fcis-4086 92 3 based base VERB fcis-4086 92 4 text text NOUN fcis-4086 92 5 classification classification NOUN fcis-4086 92 6 : : PUNCT fcis-4086 92 7 a a DET fcis-4086 92 8 comprehensive comprehensive ADJ fcis-4086 92 9 review[j review[j NOUN fcis-4086 92 10 ] ] PUNCT fcis-4086 92 11 . . PUNCT fcis-4086 93 1 2020 2020 NUM fcis-4086 93 2 . . PUNCT fcis-4086 94 1 [ [ X fcis-4086 94 2 2 2 NUM fcis-4086 94 3 ] ] PUNCT fcis-4086 94 4 lightrnn lightrnn NOUN fcis-4086 94 5 : : PUNCT fcis-4086 94 6 memory memory NOUN fcis-4086 94 7 and and CCONJ fcis-4086 94 8 computation computation NOUN fcis-4086 94 9 - - PUNCT fcis-4086 94 10 efficient efficient ADJ fcis-4086 94 11 recurrent recurrent NOUN fcis-4086 94 12 neural neural ADJ fcis-4086 94 13 networks[j networks[j PROPN fcis-4086 94 14 ] ] NOUN fcis-4086 94 15 . . PUNCT fcis-4086 95 1 2016 2016 NUM fcis-4086 95 2 . . PUNCT fcis-4086 96 1 [ [ X fcis-4086 96 2 3 3 X fcis-4086 96 3 ] ] X fcis-4086 96 4 kadlec kadlec PROPN fcis-4086 96 5 r r NOUN fcis-4086 96 6 , , PUNCT fcis-4086 96 7 schmid schmid PROPN fcis-4086 96 8 m m PROPN fcis-4086 96 9 , , PUNCT fcis-4086 96 10 ba ba PROPN fcis-4086 96 11 jgar jgar NOUN fcis-4086 96 12 o o PROPN fcis-4086 96 13 , , PUNCT fcis-4086 96 14 et et PROPN fcis-4086 96 15 al al PROPN fcis-4086 96 16 . . PUNCT fcis-4086 97 1 text text NOUN fcis-4086 97 2 understanding understanding NOUN fcis-4086 97 3 with with ADP fcis-4086 97 4 the the DET fcis-4086 97 5 attention attention NOUN fcis-4086 97 6 sum sum NOUN fcis-4086 97 7 reader reader NOUN fcis-4086 97 8 network[j network[j PROPN fcis-4086 97 9 ] ] PUNCT fcis-4086 97 10 . . PUNCT fcis-4086 98 1 2016 2016 NUM fcis-4086 98 2 . . PUNCT fcis-4086 99 1 [ [ X fcis-4086 99 2 4 4 X fcis-4086 99 3 ] ] X fcis-4086 99 4 tu tu PROPN fcis-4086 99 5 z z PROPN fcis-4086 99 6 , , PUNCT fcis-4086 99 7 liu liu PROPN fcis-4086 99 8 y y PROPN fcis-4086 99 9 , , PUNCT fcis-4086 99 10 shang shang PROPN fcis-4086 99 11 l l PROPN fcis-4086 99 12 , , PUNCT fcis-4086 99 13 et et PROPN fcis-4086 99 14 al al PROPN fcis-4086 99 15 . . PROPN fcis-4086 99 16 neural neural ADJ fcis-4086 99 17 machine machine NOUN fcis-4086 99 18 translation translation NOUN fcis-4086 99 19 with with ADP fcis-4086 99 20 reconstruction[j reconstruction[j PROPN fcis-4086 99 21 ] ] PUNCT fcis-4086 99 22 . . PUNCT fcis-4086 100 1 2016 2016 NUM fcis-4086 100 2 . . PUNCT fcis-4086 101 1 [ [ X fcis-4086 101 2 5 5 NUM fcis-4086 101 3 ] ] X fcis-4086 101 4 wan wan PROPN fcis-4086 101 5 , , PUNCT fcis-4086 101 6 j. j. PROPN fcis-4086 101 7 shan shan PROPN fcis-4086 101 8 , , PUNCT fcis-4086 101 9 wu wu PROPN fcis-4086 101 10 , , PUNCT fcis-4086 101 11 y. y. PROPN fcis-4086 101 12 c c X fcis-4086 101 13 .. .. PUNCT fcis-4086 101 14 a a DET fcis-4086 101 15 review review NOUN fcis-4086 101 16 of of ADP fcis-4086 101 17 research research NOUN fcis-4086 101 18 on on ADP fcis-4086 101 19 text text NOUN fcis-4086 101 20 classification classification NOUN fcis-4086 101 21 methods method NOUN fcis-4086 101 22 based base VERB fcis-4086 101 23 on on ADP fcis-4086 101 24 deep deep ADJ fcis-4086 101 25 learning[j learning[j NOUN fcis-4086 101 26 ] ] PUNCT fcis-4086 101 27 . . PUNCT fcis-4086 102 1 journal journal PROPN fcis-4086 102 2 of of ADP fcis-4086 102 3 tianjin tianjin PROPN fcis-4086 102 4 university university PROPN fcis-4086 102 5 of of ADP fcis-4086 102 6 technology,2021,37(02):41 technology,2021,37(02):41 PROPN fcis-4086 102 7 - - PUNCT fcis-4086 102 8 47 47 NUM fcis-4086 102 9 . . PUNCT fcis-4086 103 1 [ [ X fcis-4086 103 2 6 6 NUM fcis-4086 103 3 ] ] SYM fcis-4086 103 4 wu wu PROPN fcis-4086 103 5 , , PUNCT fcis-4086 103 6 longfeng longfeng PROPN fcis-4086 103 7 . . PUNCT fcis-4086 104 1 research research NOUN fcis-4086 104 2 on on ADP fcis-4086 104 3 news news NOUN fcis-4086 104 4 text text NOUN fcis-4086 104 5 classification classification NOUN fcis-4086 104 6 based base VERB fcis-4086 104 7 on on ADP fcis-4086 104 8 deep deep ADJ fcis-4086 104 9 learning[d learning[d PROPN fcis-4086 104 10 ] ] PUNCT fcis-4086 104 11 . . PUNCT fcis-4086 105 1 anhui anhui PROPN fcis-4086 105 2 university university PROPN fcis-4086 105 3 of of ADP fcis-4086 105 4 technology,2020 technology,2020 PROPN fcis-4086 105 5 . . PUNCT fcis-4086 106 1 [ [ X fcis-4086 106 2 7 7 X fcis-4086 106 3 ] ] X fcis-4086 106 4 clark clark PROPN fcis-4086 106 5 k k PROPN fcis-4086 106 6 , , PUNCT fcis-4086 106 7 luong luong PROPN fcis-4086 106 8 m m PROPN fcis-4086 106 9 t t PROPN fcis-4086 106 10 , , PUNCT fcis-4086 106 11 le le X fcis-4086 106 12 q q PROPN fcis-4086 106 13 v v PROPN fcis-4086 106 14 , , PUNCT fcis-4086 106 15 et et PROPN fcis-4086 106 16 al al PROPN fcis-4086 106 17 . . PROPN fcis-4086 106 18 electra electra PROPN fcis-4086 106 19 : : PUNCT fcis-4086 106 20 pre pre ADJ fcis-4086 106 21 - - ADJ fcis-4086 106 22 training training ADJ fcis-4086 106 23 text text NOUN fcis-4086 106 24 encoders encoder NOUN fcis-4086 106 25 as as ADP fcis-4086 106 26 discriminators discriminator NOUN fcis-4086 106 27 rather rather ADV fcis-4086 106 28 than than ADP fcis-4086 106 29 generators[j generators[j PROPN fcis-4086 106 30 ] ] PUNCT fcis-4086 106 31 . . PUNCT fcis-4086 107 1 2020 2020 NUM fcis-4086 107 2 . . PUNCT fcis-4086 108 1 [ [ X fcis-4086 108 2 8 8 NUM fcis-4086 108 3 ] ] X fcis-4086 108 4 jia jia PROPN fcis-4086 108 5 , , PUNCT fcis-4086 108 6 pengtao pengtao PROPN fcis-4086 108 7 , , PUNCT fcis-4086 108 8 sun sun PROPN fcis-4086 108 9 , , PUNCT fcis-4086 108 10 wei wei PROPN fcis-4086 108 11 . . PUNCT fcis-4086 109 1 a a DET fcis-4086 109 2 review review NOUN fcis-4086 109 3 of of ADP fcis-4086 109 4 deep deep ADJ fcis-4086 109 5 learning learning NOUN fcis-4086 109 6 - - PUNCT fcis-4086 109 7 based base VERB fcis-4086 109 8 text text NOUN fcis-4086 109 9 classification[j classification[j NOUN fcis-4086 109 10 ] ] PUNCT fcis-4086 109 11 . . PUNCT fcis-4086 110 1 computers computer NOUN fcis-4086 110 2 and and CCONJ fcis-4086 110 3 modernization,2021(07):2937 modernization,2021(07):2937 NOUN fcis-4086 110 4 . . PUNCT fcis-4086 111 1 [ [ X fcis-4086 111 2 9 9 NUM fcis-4086 111 3 ] ] PUNCT fcis-4086 111 4 wang wang PROPN fcis-4086 111 5 guihong guihong PROPN fcis-4086 111 6 . . PUNCT fcis-4086 112 1 research research NOUN fcis-4086 112 2 on on ADP fcis-4086 112 3 chinese chinese ADJ fcis-4086 112 4 text text NOUN fcis-4086 112 5 classification classification NOUN fcis-4086 112 6 based base VERB fcis-4086 112 7 on on ADP fcis-4086 112 8 deep deep ADJ fcis-4086 112 9 learning[d learning[d NOUN fcis-4086 112 10 ] ] PUNCT fcis-4086 112 11 . . PUNCT fcis-4086 113 1 xi'an xi'an PROPN fcis-4086 113 2 university university PROPN fcis-4086 113 3 of of ADP fcis-4086 113 4 technology,2020 technology,2020 PROPN fcis-4086 113 5 . . PUNCT fcis-4086 114 1 [ [ X fcis-4086 114 2 10 10 NUM fcis-4086 114 3 ] ] X fcis-4086 114 4 minaee minaee PROPN fcis-4086 114 5 s s PROPN fcis-4086 114 6 , , PUNCT fcis-4086 114 7 kalchbrenner kalchbrenner NOUN fcis-4086 114 8 n n NOUN fcis-4086 114 9 , , PUNCT fcis-4086 114 10 cambria cambria PROPN fcis-4086 114 11 e e PROPN fcis-4086 114 12 , , PUNCT fcis-4086 114 13 et et PROPN fcis-4086 114 14 al al PROPN fcis-4086 114 15 . . PUNCT fcis-4086 115 1 deep deep ADJ fcis-4086 115 2 learning learning NOUN fcis-4086 115 3 based base VERB fcis-4086 115 4 text text NOUN fcis-4086 115 5 classification classification NOUN fcis-4086 115 6 : : PUNCT fcis-4086 115 7 a a DET fcis-4086 115 8 comprehensive comprehensive ADJ fcis-4086 115 9 review[j review[j NOUN fcis-4086 115 10 ] ] PUNCT fcis-4086 115 11 . . PUNCT fcis-4086 116 1 2020 2020 NUM fcis-4086 116 2 . . PUNCT fcis-4086 117 1 [ [ X fcis-4086 117 2 11 11 NUM fcis-4086 117 3 ] ] PUNCT fcis-4086 117 4 wang wang PROPN fcis-4086 117 5 yibin yibin PROPN fcis-4086 117 6 . . PUNCT fcis-4086 118 1 deep deep ADJ fcis-4086 118 2 learning learning NOUN fcis-4086 118 3 - - PUNCT fcis-4086 118 4 based base VERB fcis-4086 118 5 news news NOUN fcis-4086 118 6 text text NOUN fcis-4086 118 7 classification classification NOUN fcis-4086 118 8 and and CCONJ fcis-4086 118 9 application[d application[d NOUN fcis-4086 118 10 ] ] PUNCT fcis-4086 118 11 . . PUNCT fcis-4086 119 1 chongqing chongqe VERB fcis-4086 119 2 normal normal ADJ fcis-4086 119 3 university,2020 university,2020 NOUN fcis-4086 119 4 . . PUNCT