Asian Journal of Social Sciences and Management Studies ISSN: 2313-7401 Vol. 3, No. 3, 173-181, 2016 http://www.asianonlinejournals.com/index.php/AJSSMS 173 Analysis of Life Context of On-Line Group-Buying Population by Dynamic Decision Shen-Tsu Wang1 1 Department of Commerce Automation and Management, National Pingtung University Taiwan, Province of China Abstract While it is difficult to avoid uncertainties when shopping on the Internet, trust can reduce customers’ perceived uncertainties, and enhance their willingness and frequency to buy products and services. The difference in time and space information transparency between customers and on-line sellers, as well as the complex unpredictability of network structure, result in frequent uncertainty for on-line transactions. Therefore, through text mining and integrating the Genetic Algorithm (GA) with the Support Vector Machine (SVM), this project classifies the data of on-line group buying community complaints according to the posts left on Facebook and the three major group-buying websites of Taiwan. The terms are selected based on term frequency, document frequency, uniformity, and conformity, while document classification effectiveness is calculated using precision, recall rate, and F-measure. Community complaints are classified into the uncertain performance indicators that influence on-line group buying for integrated statistics, in order that specific performance indicators of community group-buying websites can be generated. Afterwards, based on the on-line group buying community performance indicator sequence, as integrated according to the dynamic Multicriteria Optimization and Compromise Solution (VIKOR) method and prosperity countermeasure signals, grey correlation sorting is applied to analyze the dynamic performance indicator sequence of different communities, in order to determine the life context of different populations for the reference of on-line group buying providers. Keywords: Genetic algorithm, Support vector machine, On-line group buying, Life context, Grey correlation sorting, Dynamic performance indicator. Contents 1. Introduction ....................................................................................................................................................................... 174 2. Research Method ............................................................................................................................................................... 174 3. Results Analysis ................................................................................................................................................................. 180 4. Conclusion .......................................................................................................................................................................... 180 References .............................................................................................................................................................................. 180 Citation | Shen-Tsu Wang (2016). Analysis of Life Context of On-Line Group-Buying Population by Dynamic Decision. Asian Journal of Social Sciences and Management Studies, 3(3): 173-181. DOI: 10.20448/journal.500/2016.3.3/500.3.173.181 ISSN(E) : 2313-7401 ISSN(P) : 2518-0096 Licensed: This work is licensed under a Creative Commons Attribution 3.0 License Funding: This study received no specific financial support. Competing Interests: The author declares that there are no conflicts of interests regarding the publication of this paper. Transparency: The author confirms that the manuscript is an honest, accurate, and transparent account of the study was reported; that no vital features of the study have been omitted; and that any discrepancies from the study as planned have been explained. Ethical: This study follows all ethical practices during writing. History: Received: 3 February 2016/ Revised: 25 February 2016/ Accepted: 29 February 2016/ Published: 2 March 2016 Publisher: Asian Online Journal Publishing Group http://creativecommons.org/licenses/by/3.0/ http://crossmark.crossref.org/dialog/?doi=10.20448/journal.500/2016.3.3/500.3.173.181 http://search.crossref.org/?q=10.20448/journal.500/2016.3.3/500.3.173.181 http://crossmark.crossref.org/dialog/?doi=10.20448/journal.500/2016.3.3/500.3.173.181 http://search.crossref.org/?q=10.20448/journal.500/2016.3.3/500.3.173.181 Asian Journal of Social Sciences and Management Studies, 2016, 3(3):173-181 174 1. Introduction The course of the real world and contexts of places of existence are the details of human life. The theory of humanistic geography opposes absolute objective and mechanized intuitionistic humanity, and places stress on the relative meanings, different values, and objectives of persons throughout the world. Since the theory of humanistic geography pays great attention to the context of humanistic life, it is attracted and used by the theoretical concept of phenomenology, and its interesting subjects are examined, such as correlated place, mutual space, and life area (Wen, 2005). Scholars who think based on context not only consider the single development of users and the actual nature of products, they also emphasize interactions, constructed cultural aspects, and process values, and pay attention to the common structural relationship between multiangular sociocultural values and technology (Dulk et al., 2013; Trefalt et al., 2013). The uncertainties of life context include the constraints of innovation sources, as well as the differences between manufacturing and customer sides. The important findings regarding the constraints of innovation sources show that, the more accustomed individuals are to observing or using a tool in a particular way, the harder it is to escape the existing constraint architecture, and the harder it is to have innovative views of substantial content, thus, research processes are unlikely to propose novel views (Hippel, 2011). While it is difficult to avoid uncertainties when shopping on the Internet, trust can reduce customers’ perceived uncertainties, and enhance their willingness and frequency to buy products and services. The difference in time and space information transparency between customers and on-line sellers, as well as the complex unpredictability of network structure, result in frequent uncertainty for on-line transactions. Therefore, through text mining and integrating the Genetic Algorithm (GA) with the Support Vector Machine (SVM), this project classifies the data of on-line group buying community complaints according to the posts left on Facebook and the three major group- buying websites of Taiwan. The terms are selected based on term frequency, document frequency, uniformity, and conformity, while document classification effectiveness is calculated using precision, recall rate, and F-measure. Community complaints are classified into the uncertain performance indicators that influence on-line group buying for integrated statistics, in order that specific performance indicators of community group-buying websites can be generated. Afterwards, based on the on-line group buying community performance indicator sequence, as integrated according to the dynamic Multicriteria Optimization and Compromise Solution (VIKOR) method and prosperity countermeasure signals, grey correlation sorting is applied to analyze the dynamic performance indicator sequence of different communities, in order to determine the life context of different populations for the reference of on-line group buying providers (Chen et al., 2010; Cheng and Huang, 2013; Wang et al., 2013; Liang et al., 2014; Zhou and Xie, 2014; Hsu et al., 2015). 2. Research Method This project uses the Chinese Knowledge Information Processing (CKIP) of Academia Sinica to analyze the word segmentation of Chinese documents, and screens out the terms of on-line group buying community complaints. Afterwards, the optimal term combination is selected by GA to train the SVM of the existing types of documents, and the classificatory documents are classified by test file; then all other documents not classified into the existing types are clustered, and document clustering is optimized using GA (Lo et al., 2008; Trstenjak et al., 2014; Erra et al., 2015). There are two major steps, as follows. 2.1. Document Pre-Processing The document classification of this project can be divided into two parts, the original type of document cluster and new document cluster. The existing type of document cluster contains the existing architecture type of original documents, while the new document cluster contains partial existing type of documents, as well as any documents not belonging to the original type. As there is no appropriate decision-making approach for terms in Chinese documents, word segmentation is required in order to recognize all terms in the documents. Among the plans for Chinese word segmentation, as constructed by the Chinese Knowledge Information Processing Group (CKIPG) of Academia Sinica, CKIP is representative and practical. Therefore, this project uses CKIP, as provided by CKIPG, to process the training documents and test documents of this project. 2.2. GA-SVM Model This project combines GA with SVM to construct the GA-SVM computational model, in order to determine better term combinations to train the existing type of classification model, and whether or not new documents can be classified into the original type is judged according to this architecture. In order to effectively determine better term combinations, important terms are selected using GA, where the Term Frequency-Inverse Document Frequency (TF- IDF) values are calculated according to the selected terms (Lo et al., 2008; Trstenjak et al., 2014; Erra et al., 2015) as the basis of training documents and create the vector values of the documents, and the obtained document vectors are used as the input vector values of SVM to train the GA-SVM model base. The term combinations, as selected by this model, correspond to the contents of new documents in order to calculate the TF-IDF-based document vector matrices of the new documents, and the document vector matrices of the new documents are converted into the SVM format and imported into the SVM training model for classification. When new documents fail to be classified as the original type, they are extracted for second stage processing. However, when GA is selected, as it can implement efficient global search for a better solution to the problem, which can reduce the probability of obtaining a local optimal. Where TF-IDF is expressed as Eq. (1).           , , , log i i i i i TF IDF W M TF W M IDF W D TF W M DF W            (1) Asian Journal of Social Sciences and Management Studies, 2016, 3(3):173-181 175 Where,  ,iTF IDF W M : Weight of term iW in M D : Total number of messages in document  iDF W : Occurrence frequency of term iW in document set The message can represent the matrix M formed by the vector of a term, expressed as Eq. (2). 11 12 1 21 22 2 1 2 p p m m mp w w w w w w M w w w             (2) where, ikW : Weight of term k in document  1...i i m m D 1,...,k p p : Number of terms extracted from all messages Therefore, this project intends to use the characteristics of GA to determine the most representative terms to optimize the SVM model, in order to classify the complaints of on-line group-buying communities. The execution flow of the GA-SVM classifier is introduced below. This project uses the four functions of Chou et al. (2007) as the basis for selecting terms, the term frequency, document frequency, uniformity, and conformity, described as follows: 1) Select term (1) Term frequency (TF) TF represents the term occurrence probability of term i in jK type, expressed as Eqs. (3)–(4): 1 ij ij J ij j T F TF T F      (3) 1 ij ij I ij i a T F a     (4) where, i : Term j : Type jK ija : Occurrence number of term i in jK type ijT F  : Occurrence probability of term i in jK type If the occurrence number of a term i in jK type is greater than that in other types, and the calculated TF is large; this term i can represent type jK . (2) Document frequency (DF) DF represents the occurrence probability in documents of term i in jK type, expressed as Eqs. (5)–(6): 1 ij ij J ij j D F DF D F      (5) ij ij j S D F S   (6) where, ijS : Occurrence number of documents with term i in jK type ijD F  : Occurrence probability of documents with term i in jK type If the number of documents with term i in jK type is greater than the number of documents in other types, then term i better represents jK type. (3) Uniformity Uniformity represents the occurrence probability of term i in all documents of jK type, expressed as Eqs. (7)– (8): 1 log jS ij ic ic c U b b    (7) Asian Journal of Social Sciences and Management Studies, 2016, 3(3):173-181 176 1 j ic ic S ic c tf b tf    (8) where, c : Document jS : Total number of documents within jK type ictf : Occurrence number of term i in c documents icb : Average occurrence probability of term i in document c If the occurrence number of term i in jK type documents is greater than the occurrence number of other terms in jK type documents, the larger the calculated uniformity, the better this term i represents this type than other terms. (4) Conformity The conformity represents the probability of the occurrence of term i in all types, expressed as Eqs. (9)–(10): 1 log J i ij ij j CF P P    (9) 1 ij ij J ij j S P S    (10) where, ijP : Probability of occurrence of document S with term i in j type When term i occurs in all types, the iCF value is large. Contrarily, when i only occurs in one single type, the iCF value is 0. Therefore, the smaller the iCF , the better the term i represents the type. 2) Classifier training by SVM Each term will obtain four threshold values. This project selects the terms meeting the four thresholds as the basis of training documents, then the TF-IDF value of the obtained term is calculated, and the document vector value is created for each document as the input vector of SVM for the training classifier. (1) Model parameter selection This project uses a library for support vector machines (LIBSVM) (Chang and Lin, 2001) for document classification, as LIBSVM provides multiple parameter settings, four core functions, four classifiers, and an important parameter search tool, thus, enabling the user to adjust parameters according to different problems in order to increase the efficiency of classification. However, as document classification is mostly a polytype classification problem, this project uses the C-Support Vector Classification (C-SVC) form, which supports the multi-classification model to train SVM. The terms meeting the threshold set value are extracted, the document vector is created, and the vector value is imported into LIBSVM to train the SVM classifier. (2) Predict training data by SVM When the SVM model of some chromosomes is generated, the SVM classification model can be used to predict the classification of training documents. As this project uses GA for continuous calculation to determine the representative term combination, the condition of calculation is to calculate the fitness value according to the mean F-measure of SVM. Afterwards, if the calculation reaches the stop condition, and the SVM has good classification effectiveness, we can use this classification model to classify new community complaint documents. 3) GA-SVM fitness function design In the course of document classification, document classification effectiveness is calculated based on Precision, Recall, and F-measure. The fitness function of this project uses the F-measure mean as the basis of evaluating classification. Therefore, the larger the fitness function value, the better the extracted term classifies the document into the original type, where the fitness function is expressed as Eqs. (11)–(14): , , 1 i I s p K i F fitness I   (11) F-measure: , , , , , , , , , , 2 i i i i i s p K s p K s p K s p K s p K P R F P R    (12) Precision: , , , , , i i i i p K s K s p K s K N N P N   (13) Recall: , , , , , i i i i p K s K s p K p K N N R N   (14) where, i : type iK Asian Journal of Social Sciences and Management Studies, 2016, 3(3):173-181 177 , is KN : Number of documents classified as iK by classifier , ip KN : Number of documents defined as iK by training document , ,i ip K s KN N : Number of documents classified as iK by training document and classifier The aforesaid analyzed on-line group buying complaints are put in the classification influencing on-line group buying community uncertainty in Table 1 and calculated. Table-1. Analysis of performance indicators influencing on-line group buying community uncertainty Group-buying website performance aspect Group-buying website performance indicator Operational definition A. Community word-of-mouth acceptance A1. Consumption experience Community experience in buying or using the goods. A2. Specialty Community word-of-mouth acceptor's awareness of the merchandise information. B. Community word-of-mouth content B1. Positive/negative word-of-mouth information Positive/negative word-of-mouth type of commodity to be bought by community, including fully positive word-of-mouth, fully negative word-of-mouth, and half and half positive/negative word-of-mouth. B2. Information quantity Number of word-of-mouth comments on the commodity to be bought by community. C. Master buyer reliability C1. Reputation and evaluation Whether the past evaluation of the master buyer is good, and whether the master buyer is reliable. C2. Friendly interaction Master buyer responds to members' questions, and informs members of the matters concerned with group buying. D. Community perceived risk D1. Financial risk Probable monetary loss of community from on-line shopping. D2. Social Risk Whether the community's shopping on the website is objected by peers. D3. Psychological risk Community has negative opinion after shopping or repents buying. D4. Physical Risk Using commodity causes physical injury. D5. Performance risk Whether the function of a commodity bought from the internet meets the original expectations of the community. D6. Time risk Time spent searching for the desired commodity online and obtaining the commodity. D7. Privacy risk Identity data registered for on-line transactions are illegally used by others. E. Community security E1. Disturbance-free shopping environment Take part in on-line group buying for disturbance-free shopping environment. E2. Trust group-buying website Take part in on-line group buying for trusting master buyer or group-buying website. E3. Group-buying products are relatively quality Take part in on-line group buying, as group-buying commodities have relative quality assurance. F. On-line group buying diversity F1. Novel commodities Take part in on-line group buying for extraordinary commodities that are unavailable on the market. G. On-line group buying price G1. Lower price Take part in on-line group buying for buying more commodities at lower prices. G2. Free gift Take part in on-line group buying for gifts. H. On-line group buying convenience H1. Convenient receiving Take part in on-line group buying for convenient receiving and delivery services. I. Website design I1. Clear information presentation Take part in on-line group buying for explicit website information. I2. Simple shopping process Take part in on-line group buying for simple shopping process. Source: (Hsu et al., 2014; Nepomuceno et al., 2014; See-To and Ho, 2014; Zhao and Zhu, 2014) 2.3. Dynamic Performance Indicator Sequencing - Integrate Grey Correlation Sorting with Dynamic VIKOR Asian Journal of Social Sciences and Management Studies, 2016, 3(3):173-181 178 Considering the prosperity countermeasure signal, the performance indicator influencing the uncertainty on-line group buying community is analyzed in Table 1. The prosperity countermeasure signal uses five signal lights to represent a situation of prosperity. The cross-check scores are shown in the brackets following the light signal (Ou- Yang and Chuang, 2007) (1) red light (38-45 points) represents overheating; (2) yellow-red light (32-37 points) represents activated prosperity; (3) green light (23-31 points) represents stable prosperity; (4) yellow-blue light (17- 22 points) represents poor prosperity; (5) blue light (9-16 points) represents business recession (Deng, 1982; Golmohammadi and Mellat-Parast, 2012; Zhu and Hipel, 2012; Chien, 2015). The evaluation process of this project follows the dynamic VIKOR multicriterion evaluation method, where iw considers the prosperity countermeasure signal, and performance is evaluated according to the weights of various phases and sequencing processes (Hung et al., 2013). Step1: Evaluate the scores of ratees in various periods If each reviewer uses a triangular intuitionistic fuzzy number to represent the result of the criteria, according to the selection criteria of ratees, the denotation is  , , ; ,a b c x y , where , ,a b c represent the fuzzy value of a single criterion result of a ratee, the triangular fuzzy number display is flexible, ,x y represent the degree of consent and objection of the decision maker, respectively, and the best result of the assessment score (ideal solution) is represented by  10,10,10;1,0 . Step 2: Calculate the distance between each ratee's criteria results and the best score According to Szmidt and Kacprzyk (2000) when the distance between the criteria  1,2,...,iD i n result in each phase  1,2,...,kT k p of each ratee  1,2,...,jD j m and the optimal value is calculated, the distance matrix   ~ 1,2,...,kG k p of various phases can be obtained. Step 3: Evaluation by the VIKOR multicriterion evaluation method (1) Determine the best value and worst value in the criterion function Each ratee  1,2,...,jB j m  is evaluated under evaluation criteria  1,2,...,iD i n , under No. iD evaluation criterion, the performance evaluation value of alternative jB is represented by ijR , the best value * iR and worst value iR of criterion function vary with the effectiveness (larger-the-better) criterion and cost (smaller-the-better) criterion, expressed as Eqs. (15)–(16). If there are m alternatives  1,2,...,jB B j m  , the evaluation is implemented under n evaluation criteria  1,2,...,iD D i n  , under No. i evaluation criterion, the performance evaluation value of alternative jB is represented by ijR . If the criterion is the effectiveness (larger-the-better) criterion, the best value * iR and worst value iR are expressed as Eq. (15). * ax ,i ij i ij j j R M R R Min R  (15) On the contrary, if the criterion is the cost (smaller-the-better) criterion, the best value * iR and worst value iR are expressed as Eq. (16). * ,i ij i ij j j R Min R R Max R  (16) (2) Normalization When the distance between the triangular intuitionistic fuzzy result value of each ratee under the selection criteria, and the best result  10,10,10;1,0 is obtained, the normalized value of ratees under criteria is calculated by Eq. (17). The normalized value of jB scheme under iD criterion is represented by  i jV B , expressed as Eq. (17):   * * i ij i j i i R R V B R R    (17) (3) Calculate jA and jC values The criterion normalized value of a ratee in each phase is multiplied by the weight of the criterion, according to Eq. (18). Finally, the weighted normalized values are summed up to obtain jA , and the criterion normalized weights of ratees are compared according to Eq. (18) where the maximum value is jC , expressed as Eq. (19).     * * 1 , n i ij j i i i i R R A w j R R       (18)     * * , i ij j i i i i R R C Max w j R R         (19) (4) Select * *, , ,A A C C  values Asian Journal of Social Sciences and Management Studies, 2016, 3(3):173-181 179 When jA and jC are obtained, *,A A are selected from jA of criteria, *A represents the minimum jA value  * j j A Min A , A represents the maximum jA value  j j A Max A  ; *,C C are selected from jC of criteria, *C represents the minimum jC value  * j j C MinC , C  represents the maximum jC value  j j C MaxC  . (5) Calculate jU value The jU value of each scheme is calculated according to Eq. (20). The decision mechanism coefficient  (weight) is set as 0.5, in order to simultaneously maximize group utility and minimize specific regret.           * * * * 1 , j j j A A C C U j A A C C                         (20) Step 4: Implement weighting and sequencing according to the weights of various periods Chen and Li (2011) proposed the multiple attribute decision making problem, with m decision making sides  1,2,...,jB j m and n evaluation attributes  1,2,...,iD i n , where the weight  1,2,...,iw n of each attribute should be considered, and the probably of different performances in multiple phases  kT must be considered, in order to perfect the evaluation results of various proposals. In this project, as the jU value is smaller-the-better criterion (smaller is better), polarity inversion is implemented for the weight of period, and the converted actual weight is represented by  * 1,2,...,tkw k p , expressed as Eq. (21):     * 1 1 1 tk tk p tk k w w w     (21) The jU value of various phases is represented by  1,2,...,jkU k p , and multiplied by actual weight  * 1,2,...,tkW k p , in order to obtain the actual weight *jkU of each period, expressed as Eq. (22). * *jk tk jkU w U  (22) The weighted *jkU values of the phases are summed up to obtain the T value, expressed as Eq. (23): * 1 p jk k T U   (23) When the smaller weighted sum T of the *jkU value of various periods is better, the total sequence of the full phase can be obtained. This project is based on dynamic VIKOR, with the prosperity countermeasure signal and analysis results as shown in Table 1. The degree of relationship among sub-systems or elements could be evaluated through grey relational analysis, and the important influential factors of the development trend are determined in order to learn the major features of the system through the following steps, as shown in Eq. (24)-(29) (Deng, 1982; Chien, 2015). Step 1: Normalize original data. Normalize by dividing the original data ( )ix k with the mean value of the sequence shown in Eq. (1): 1 ( ) ( ) , ,..., ,..., ( ) i i N i k x k r k i a d k A N x k N     (24) Step 2: Designate the standard sequence and calculate the difference sequence. Take the mean value as a standard sequence, i.e. sequence 0; the difference sequence 0 ( )i k indicates the absolute difference of elements k between the other sequence i and the standard sequence 0, as expressed in Eq. (5): 0 0( ) ( ) ( ) , 1,2,3,... ,...,i ik r k r k i k A N     (25) Step 3: Calculate maximal difference max and minimal difference min , as expressed in Eqs. (6) and (7). max 0 , ( )i i k Max k   (26) min 0 , ( )i i k Min k   (27) Step 4: Calculate grey relational coefficient: 0 ( )i k . The relational coefficient: 0 ( )i k is defined below, of which  is the adjustment factor, as shown in Eq. (8). min max 0 0 max ( ) ( ) i i k k           (28) Step 5: Calculate the grey relationship 0i between each sequence and the standard sequence. The grey relationship 0i is defined as in Eq. (9). Asian Journal of Social Sciences and Management Studies, 2016, 3(3):173-181 180 0 0 ( )N i i k A k N     (29) Step 6: Conduct sequencing according to the grey relationship. 3. Results Analysis According to the 863 data collected from group-buying websites, as shown in Table 2 the military, civil servants, and teachers account for a high proportion of group buying. The fitness values of various populations, as calculated by Eq. (11) are greater than 95%, meaning the extracted term is well effective at classifying the document as the original type. The performance indicators of different populations are calculated using the dynamic VIKOR method. Afterwards, the key performance indicator of each population is calculated by grey correlation sorting. Students and homemakers care most about minimum price; the freelance and service industry care most about convenient receiving; the service industry cares most about simple shopping processes; the financial services/insurance industry cares most about financial risk; the military, civil servants, and teachers care most about reliable group-buying websites; the information industry cares most about the master buyer's reputation and evaluation. The analysis results fully show the life contexts of different populations, as well as the on-line group buying content they are most interested in. Table-2. Group buying population classification Population type Proportion Fitness T(value, Performance indicator) 0i (Value, Key performance indicator) Students 4.53% 96.11% 0.26 G1,G2,F1 0.56 G1 Homemakers 13.26% 97.18% 0.13 G1,E2,C2 0.58 G1 Freelance 12.25% 96.21% 0.18 A2,H1,E2 0.66 H1 Manufacturing industry 6.26% 95.13% 0.11 E2,H1,I2 0.71 H1 Service industry 16.38% 98.32% 0.28 E2,I2,I1 0.68 I2 Financial services / insurance industry 8.25% 95.29% 0.19 G2,D2,D1 0.79 D1 Military, civil servants and teachers 28.69% 97.28% 0.09 E2,D2,D1 0.87 E2 Information industry 10.38% 95.16% 0.27 D2,D1,C1 0.81 C1 Source: This research 4. Conclusion The life contexts and habits of different populations of on-line group buying match the results of this study; student and homemakers care most about minimum price, because the economic situation of this population is relatively controlled; the freelance and service industry care most about convenient receiving, because their work places are inconvenient for receiving; the service industry cares most about simple shopping processes, because they have less spare time; the financial services/insurance industry cares most about financial risk, due to their profession; the military, civil servants, and teachers care most about reliable group-buying websites, due to their cautious behavior style; the information industry cares most about master buyer's reputation and evaluation, due to sensitivity to information. References Chang, C. and C. Lin, 2001. LIBSVM: A library for support vector machines. Available from https://www.csie.ntu.edu.tw/~cjlin/libsvm/ [Accessed January 20 2015]. Chen, J., R.J. Kauffman, Y. Liu and X. Song, 2010. Segmenting uncertain demand in group-buying auctions. Electronic Commerce Research and Applications, 9(2): 126-147. Chen, Y. and B. Li, 2011. Dynamic multi-attribute decision making model based on triangular intuitionistic fuzzy numbers. Scientia Iranica, 18(2): 268-274. Cheng, H.H. and S.W. Huang, 2013. Exploring antecedents and consequence of online group-buying intention: An extended perspective on theory of planned behavior. International Journal of Information Management, 31(1): 185-198. Chien, C.F., 2015. Decision analysis and management: A unison framework for total decision quality enhancement. Taipei: Yeh Yeh Book Gallery. Chou, C.H., C.C. Han and Y.H. Chen, 2007. GA based optimal keyword extraction in an automatic Chinese web document classification system. Lecture Notes in Computer Science, 47(43): 224-234. Deng, J.L., 1982. Control problems of grey systems. System and Control Letters, 1(5): 288-294. Dulk, L.D., S. Groeneveld, A. Ollier-Malaterre and M. Valcour, 2013. National context in work-life research: A multi-level cross-national analysis of the adoption of workplace work-life arrangements in Europe. European Management Journal, 31(5): 478-494. Erra, U., S. Senatore, F. Minnella and G. Caggianese, 2015. Approximate TF–IDF based on topic extraction from massive message stream using the GPU. Information Sciences, 292: 143-161. Golmohammadi, D. and M. Mellat-Parast, 2012. Developing a grey-based decision-making model for supplier selection. International Journal of Production Economics, 137(2): 191-200. Hippel, E.V., 2011. Customer is your innovation director. Taipei City: Goodness Publishing House. Hsu, M.H., C.M. Chang, K.K. Chu and Y.J. Lee, 2014. Determinants of repurchase intention in online group-buying: The perspectives of DeLone & McLean IS success model and trust. Computers in Human Behavior, 36: 234-245. Hsu, M.H., C.M. Chang and L.W. Chuang, 2015. Understanding the determinants of online repeat purchase intention and moderating role of habit: The case of online group-buying in Taiwan. International Journal of Information Management, 35(1): 45-56. Hung, K.C., C.T. Chang and K.L. Chu, 2013. Developing dynamic multiple criteria decision making model for military officer performance appraisal. Journal of National Defense Management, 34(1): 17-27. Liang, X., L. Ma, L. Xie and H. Yan, 2014. The informational aspect of the group-buying mechanism. European Journal of Operational Research, 234(1): 331-340. Lo, S.C., S.K. Ke and C.C. Lin, 2008. Web-service data quality mining and control. Chiao Da Management Review, 28(1): 251 - 268. http://www.csie.ntu.edu.tw/~cjlin/libsvm/ Asian Journal of Social Sciences and Management Studies, 2016, 3(3):173-181 181 Nepomuceno, M.V., M. Laroche and M.O. Richard, 2014. How to reduce perceived risk when buying online: The interactions between intangibility, product knowledge, brand familiarity, privacy and security concerns. Journal of Retailing and Consumer Services, 21(4): 619-629. Ou-Yang, H. and S.S. Chuang, 2007. An empirical study on the relationship among corporate governance, relative performance evaluation, and top executive turnover. Asia-Pacific Economic and Management Review, 11(1): 53-86. See-To, E.W.K. and K.K.W. Ho, 2014. Value co-creation and purchase intention in social network sites: The role of electronic word-of-mouth and trust – a theoretical analysis. Computers in Human Behavior, 31: 182-189. Szmidt, E. and J. Kacprzyk, 2000. Distances between intuitionistic fuzzy sets. Fuzzy Sets and Systems, 114(3): 505-518. Trefalt, Š., M. Drnovšek, A. Svetina-Nabergoj and R.V. Adlešič, 2013. Work-life experiences in rapidly changing national contexts: Structural misalignment, comparisons and choice overload as explanatory mechanisms. European Management Journal, 31(5): 448-463. Trstenjak, B., S. Mikac and D. Donko, 2014. KNN with TF-IDF based framework for text categorization. Procedia Engineering, 69: 1356- 1364. Wang, J.J., X. Zhao and J.J. Li, 2013. Group buying: A strategic form of consumer collective. Journal of Retailing, 89(3): 338-351. Wen, J.K., 2005. Humanity psychiatry is beneficial for psychosocial rehabilitation of the mentally ill. Taiwanese Journal of Psychiatry, 19(2): 83-84. Zhao, B. and Y. Zhu, 2014. Formalizing and validating the web quality model for web source quality evaluation. Expert Systems with Applications, 41(7): 3306-3312. Zhou, Y. and J. Xie, 2014. Potentially self-defeating: Group buying in a two-tier supply chain. Omega, 49: 42-52. Zhu, J. and K.W. Hipel, 2012. Multiple stages grey target decision making method with incomplete weight based on multi- granularity linguistic label. Information Sciences, 212: 15-32. Asian Online Journal Publishing Group is not responsible or answerable for any loss, damage or liability, etc. caused in relation to/arising out of the use of the content. Any queries should be directed to the corresponding author of the article.