INTERNATIONAL JOURNAL OF COMPUTERS COMMUNICATIONS & CONTROL
ISSN 1841-9836, 12(2):183-200, April 2017.

Model of Network Topic Detection Based on
Web Usage Behaviour Mode Analysis and Mining Technology

M. Chen

Mo Chen
1. Business College of Beijing Union University
A3, Yanjingdongli, Chaoyang district, Beijing, 100025, P.R. China.
mo.chen@buu.edu.cn
2. School of Information, Renmin University of China.
No.59, Zhongguancun Street, Haidian District, Beijing, 100872, P.R. China.
chenmoky@sohu.com

Abstract: This research has caught researchers’ wide attention for detecting network
topic exactly with the arrival of big data era characterized by semi-structured or un-
structured text. This paper proposes a model of network topic detection based on web
usage behaviour mode analysis and mining technology taking Web news as object of
research. The author elaborates main function and method proposed in this model,
which include the analysis module of Web news instance clicking mode, the analysis
module of Web news instance retrieval mode, the analysis module of Web news in-
stance seed and the analysis module of similar Web news instance supporting topics.
Based on these functions and methods, the author elaborates main algorithm pro-
posed in this model, which include the mining algorithm of Web news seed instances
and the mining algorithm of similar Web news instances supporting topics. These
functional algorithms have been applied in processing module of model, and focus on
how to detect network topic efficiently from a large number of web usage behaviour
towards to Web news instances, in order to explore a research method for network
topic detection. The process of experimental analysis includes three steps, firstly,
the author analyses the precision of topic detection under different method, secondly,
the author completes the impact analysis of Web news topic detection quality from
the number of Web news instances concerned and seed threshold, finally, the author
completes the quality impact analysis of Web news instances mined supporting topic
from the number of Web news instances concerned and probability threshold. The
results of experimental analysis show the feasibility, validity and superiority of model
design and play an important role in constructing topic-focused Web news corpus so
as to provide a real-time data source for topic evolution tracking.
Keywords: web usage behaviour, network topic detection, clicking mode analysis,
retrieval mode analysis.

1 Introduction

With the arrival of big data era, the field of information technology and Internet has developed
a challenging stage so far. According to survey of TeckTarget that is a global leading professional
IT network media [15] [2] [22], it has shown that the number of enterprises’ data has broken
through PB level with development of network, social media, business and other fields. Based on
data existed and existing, people should think how to analyse complicated network data showing
a tendency of explosive growth [4] [5], which have been concerned and are characterized by
semi-structured or unstructured text, nevertheless, in whole process of cognizing network data,
detecting topic exactly and effectively is the important and critical application direction.

In a mass of network data, the number of Web news released has reached EB level with events
that continue to take place in social [6] [7], which shows the 4V features of big data, it is volume,
variety, velocity and value [15] [9]. Based on these features above, Web news should reflect

Copyright © 2006-2017 by CCC Publications



184 M. Chen

high currency and reliability, on the basis of which, the topic contained in Web news should be
condensed quickly and its path of evolution should also be tracked nearly in real time. However,
how to detect network topic efficiently from a large number of web usage behaviour towards to
Web news instances, it has become an urgent problem solved to construct a topic-focused Web
news corpus so as to provide real-time data source for topic evolution tracking.

This paper proposes a model of network topic detection mainly containing four processing
modules based on web usage behaviour mode analysis and mining technology taking Web news as
object of research. The author elaborates function, method and technology on every processing
module of the model in detail, which have been used or completed, and focuses on how to detect
network topic efficiently from massive web usage behaviour towards to Web news instances. This
process of research does key contribution for exploring a method for network topic detection, this
experimental analysis results show the feasibility, validity and superiority of model design and
implement.

2 Related works

In recent several years, some scholars have conducted some research about network topic de-
tection method using different theory and technology. For example, Yang et all. survey research
on the method of topic link detection based on improved information bottleneck theory [10], in
this paper, a method of representing text is proposed, which can divide text into several sec-
tions of sub-topic features based on the regular pattern of semantic distribution and improve
information bottleneck theory, then, the text represented by the attributes is utilized to do topic
link detection, the experimental results have shown that this method has a fast convergent rate,
and can improve the performance of topic link detection system. Suhara, Yoshihiko and others
survey research on the method of information detection based on sentence-level topic [11], in
this paper, the text sentence-level diversity features based on the probabilistic topic model is
proposed, an information content classifier is also constructed combining features proposed, the
experimental results show that this method outperforms the conventional methods. Pang, JB
and others survey research on the method of unsupervised web topic detection using a ranked
clustering-like pattern across similarity cascades [12], in this paper, a method using a clustering-
like pattern across similarity cascades is investigated from the perspective of similarity diffusion,
a topic-restricted similarity diffusion process is also proposed to identify real topic from a large
number of candidates efficiently, the experimental results demonstrate that this approach out-
performs the state-of-the-art methods on several public data sets, those works are related to
author’s research direction of network topic detection and application.

In recent several years, some scholars have also conducted certain research about method
and technology of web usage behaviour analysis and mining. For example, Dziczkowski, Grze-
gorz and others survey research on the opinion mining approach for web user identification and
clients’ behaviour analysis [13], in this paper, an approach based on statistical analysis of natural
language is proposed, three different methods are used for classifying opinions from clients’ data,
two new methods are introduced based on linguistic knowledge, in order to assign a mark depen-
dent upon the client’s emotions and opinions described in comments, the effect of experiments
demonstrates that the system developed can carry out an evaluation and rating of opinions.
Karakostas, Bill and others survey research on the MapReduce architecture for web site user be-
haviour monitoring in real time [14], in this paper, a MapReduce style architecture is proposed,
where the processing of event series from the web users is performed by a number of cascading
mappers, reducers, local to the event origin, the experimental results show that this architecture
is capable to carry out time series analysis in real time for very large web data sets based on
the actual events instead of resorting to sampling or other extrapolation techniques. Zhang,



Model of Network Topic Detection Based on
Web Usage Behaviour Mode Analysis and Mining Technology 185

YH and others survey research on the new replacement algorithm of web search based on user
behaviour [15], this paper analyses the search ranking based on the user behaviour investigated
mass distribution of information on the website, then proposes a replacement algorithm for web
search, the simulation experimental results show that this approach under the search algorithm
can reduce the execution time of retrieval effectively, and the optimal parameter selection for this
blocking organization can be discussed continuously, those works are related to author’s research
direction of web usage behaviour application for analysis and mining.

Based on the analysis of the related research on network topic detection method and web
usage behaviour application technology, experts and scholars have studied on two directions,
but the research of constructing a network topic detection model based on web usage behaviour
mode analysis and mining technology taking Web news as an object of analysis according to its
attention and usage trait is missing. Therefore, this paper proposes a model of network topic
detection based on web usage behaviour mode analysis and mining technology mainly, in order
to explore how to detect network topic accurately.

3 Problem definition and notations

With rapid development of information and network technology, there are many types of
network information, such as short text of micro blog, short, moderate or long text of Web news,
long text of document and so on, while the biggest difference is structure of text content among
them. This paper selects Web news as the object of research in view of ensuring high adaptability
that the model of network topic detection based on web usage behaviour mode analysis and
mining technology should have, in order to achieve the ideal effect of topic detection in the
aspects of analysis precision and so on, this research provides scientific method for constructing
and validating model of network topic detection.

3.1 Usage feature analysis

Users can search and browse Web news from different dimensions, granularities and frequen-
cies, which have been extracted and analyzed, these processes have been elaborated in previous
journal article published by authors [16], [17], [18], [19]. In the process of searching and browsing
Web news, the user usage behaviour can be recorded, which not only explains Web news features
used by users, but also contains the concerning topics hidden in Web news instances. Therefore,
based on the analysis of Web news usage features, it is conducive to discover knowledge hidden
in massive Web news, detect topic that the users are concerned about, track a series of events
occurring in topic, and comb out process of event evolution.

From the perspective of global usage, every Web news instance concerned by users can be
seen as a node in the range of Web news websites with authority, and the node set of some
relevant instances supporting social events can be considered as a topic, each topic can also
trigger a series of events, therefore, when users directly conceren a series of topics reflected by
the social event, not only browse multi Web news instances content that support the topic, but
also browse a series of events triggered by the topic [21]. From the perspective of local usage,
when users search for Web news, in addition to input keywords that are related to the social
events reported by Web news, but also input semantic keywords that may appear in Web news
title or content, core event, core event occurring time, core event occurring location, subjective or
objective object of the core event, and relation event information triggered by the core event [21].
Therefore, in the process of topic detection towards to Web news, if the usage features of Web
news can be considered, then it can be mined for social events reported by Web news, topics



186 M. Chen

concerned by social events, Web news instances supporting topics, which will provide a Web
news topic corpus with high quality for Web news topic evolution analysis.

3.2 Topic detection norm analysis

Based on the analysis towards to the usage features of Web news, if logical relationship
need to be mined from behaviour data among the social events reported by Web News, the
topics concerned in social events, Web news instances supporting topics, then the data and
norms adopted should be specified in the process of topic detection [22], [23], which include
user behaviour records, Web news clicking frequency based on S-U, Web news clicking frequency
corresponding to URL, Web news clicking rate corresponding to URL and so on.

In the application process of Web news topic detection for social events, users can input the
interesting keywords searched, when clicking submit request, Web news page will show multi-
tudinous title, releasing time, releasing source and content link of Web news instances, and when
users click on the contents link of the Web news instance, the application platform will record
user names, the search information submitted, the behaviour clicking on Web news instances,
the usage time and other data items, in which the search information submitted is expressed in
English, but processed in this paper by the way of Chinese.

Based on the Web news usage behavior above, (s,u) of S − U information can express
retrieval keywords and Web news instances URL contained in behavior synchronously, fq(s,u)
can express Web news instances clicking frequency based on S −U, which explains the number
of (s,u) appearance in a certain time period, fqi(u) can express Web news clicking frequency
corresponding to URL, which explains the number of the Web news instance appearance in the
i particle size of a certain time period, fq(u) can express Web news instance clicking frequency
in a certain time period as shown in formula 1, rti(u) can express Web news clicking rate
corresponding to URL as shown in formula 2.

fq(u) = fq1(u) + . . . + fqi(u) + . . . + fqn(u) (1)

rti(u) =
fqi(u)

fq(u)
(2)

Based on the Web news usage behavior above, it can be converted into a graph G = (S,U,E),
in which S can express the set of retrieval keywords submitted, U can express the set of Web
news instances URL clicked, E can express the set of edges between S and U, the edge (s,u)
can express behavior of clicking the Web news instance after submitting search request for users,
whose weight value is fq(s,u) corresponding to it.

S(u) can express the set of S, which directly connect with u in G, U(s) can express the set
of U, which directly connect with s in G, D(s) can express the degree of node in search request,
which is the number of Web news instance nodes that is connected to retrieval requirement,
D(u) can express the degree of the Web news instance node, which is the number of retrieval
requirement nodes that is connected to the Web news instance as shown in figure 1, in which the
S set is expressed in English, but processed in this paper by the way of Chinese.

3.3 Problem Notations

In this section, the author provides notations used in model and algorithms based on practical
value and application direction of Web news topic detection. Let NewsSet be a set of Web news
instances, which is a data source using Web news for user and contains a large number of instances
in Web news websites with authority. Let UserBehavior be a set of records using Web news



Model of Network Topic Detection Based on
Web Usage Behaviour Mode Analysis and Mining Technology 187

Figure 1: The behavior transition diagram based on Web news usage for users

behavior for users, which is also a data source of Web news topic detection, the specific notations
are as follows.

Definition 1. Given a set of NewsSet, it can denote using NewsSet = {ns1, . . . ,nsi, . . . ,nsk}, the
range of i is between one and k.nsi.url stores address of the Web news instance, nsi.title stores
title of the Web news instance, nsi.pubtime stores releasing time of the Web news instance,
nsi.pubsource stores releasing source of the Web news instance, nsi.content stores content of
the Web news instance, nsi.keyword stores keywords of the Web news instance, the extraction
process of these information has been elaborated in previous articles published by authors [16],
[17].

Definition 2. Given a set of UserBehavior, it can denote using
UserBehavior = {ub1, . . . ,ubi, . . . ,ubn}, the range of i is between one and n.ubi.username
stores user names using Web news, ubi.searchword stores keyword retrieving Web news, ubi.url
stores URL of the Web news instances clicked by users, ubi.systemtime stores system time using
Web news for users.

Definition 3. Based on the definition and notations above, the problem that the Web news topic
detection model needs to solve is to detect topic set contained in massive Web news instances
from massive Web news usage behavior, and mine set of Web news instances that can support
relevant topic, this result can denote using TopicURL = {tu1, . . . , tui, . . . , tuk}, the range of i is
between one and k.tui =< Topic,Topicurl >, tui.Topic can express topic description detected,
tui.Topicurl can express the set of Web news instances URL mined, which can support relevant
topic detected.



188 M. Chen

4 The design of network topic detection model

In the era background of big data development, it has become an important research direction
to detect network topic exactly in Web text mining field through the process of defining detection
targets, extracting valuable network information, analysing web user usage behaviour, mining
potential topics and applying topics detected and so on.

Based on this process, the model of network topic detection based on web usage behaviour
mode analysis and mining technology taking Web news as object of research is divided into four
modules, which include the analysis module of Web news instance clicking mode, the analysis
module of Web news instance retrieval mode, the analysis module of Web news instance seed
and the analysis module of similar Web news instance supporting topics as showed in figure 2.

Figure 2: The model of network topic detection

4.1 The analysis module of Web news instance clicking mode

The inputting content of this module is set of records using Web news behaviour for users,
the outputting content of this module is analysis results of Web news instance clicking mode,
the main function of this module is to analyse the outbreak and concern mode of Web news
instances according to records using Web news behaviour for users, and infer clicking mode of
Web news instance.

Based on analysis of Web news usage behaviour, in a certain period of time, the Web news
instance mays show a clear outbreak mode regarded as a kind of sensor for social event topic
aggregation in records of user’s Web news usage behaviour, which indicates whether the Web news
instance is able to describe relevant topic reflected in social events. Therefore, in this module,
firstly, administrator should use the outbreak mode of the Web news instance to measure whether
it is able to describe corresponding topic reflected in social events, whose value is expressed with
BR(u) that is a conjecture to sharpness of clicking rate changes, and its range is between 0 and
1 as shown in formula 3, this formula cites characteristic of entropy.

BR(u) = 1 − (−(rt1(u)lognrt1(u) + . . . + rti(u)lognrti(u) + . . . + rtn(u)lognrtn(u))) (3)



Model of Network Topic Detection Based on
Web Usage Behaviour Mode Analysis and Mining Technology 189

In the above formula, n represents the number of continuous granularity components clicked
for Web news instances, the granularity can be set at hour or day for unexpected events in
society, while the granularity can be set at week or month for normal events in society, if the
fluctuations of clicking rate is not strong in the granularity setting for the Web news instance,
then BR(u) value is smaller, if the fluctuations of clicking rate is strong in the granularity setting
for the Web news instance, then BR(u) value is bigger.

Although using formula 3 can explain outbreak mode of the Web news instance, but its value
cannot be absolutely certain that the key information of the Web news instance can describe
topic in social events, because the BR(u) value may be one, while the Web news instance is only
once clicked by users. Therefore, in this module, secondly, administrator should use the concern
mode of the Web news instance to measure whether it is able to describe corresponding topic
reflected in social events again, whose value is expressed with CR(u), and its range is between 0
and 1 as shown in formula 4.

CR(u) =
log(fq(u)) −Minui∈U (log(fq(ui)))

Maxui∈U (log(fq(ui))) −Minui∈U (log(fq(ui)))
(4)

Because the process clicked has a characteristic of power law distribution for Web news
instances, so in the above formula, the clicking frequency of the Web news instance is transformed
to logarithm, if the Web news instance can be concerned by more users, then CR(u) value is
larger, whereas, CR(u) value is smaller. Based on the above measurement of outbreak and
concern mode for Web news instances, in this module, finally, administrator should use formula
5 to infer results of the Web news instance clicking mode.

ClickMode(u) = BR(u)CR(u) (5)

4.2 The analysis module of Web news instance retrieval mode

The inputting content of this module is set of records using Web news behaviour for users,
the outputting content of this module is analysis results of Web news instance retrieval mode,
the main function of this module is to analyse the degree distribution and similar mode of Web
news instances according to records using Web news behaviour for users, and infer the retrieval
mode of Web news instance.

Based on analysis of behaviour diagram G using Web news for users, it can be found that the
degree of Web news instances shows a characteristic of power law distribution, so in this module,
firstly, administrator should use the degree distribution mode of Web news instance to measure
whether it is able to describe corresponding topic reflected in social events and express it using
DR(u) as shown in formula 6.

DR(u) =
log(d(u)) −Minui∈U (log(d(ui)))

Maxui∈U (log(d(ui))) −Minui∈U (log(d(ui)))
(6)

Although using formula 6 can explain degree distribution mode of the Web news instance,
but its value cannot be absolutely certain that the key information of the Web news instance can
describe topic in social event, because the reason of generating Web news instance degree is that
users search it using keywords. Therefore, in this module, secondly, administrator should use the
similar mode of Web news instance to measure whether it is able to describe corresponding topic
reflected in social events again, whose value is expressed with SS(u), in order to solve problem
of existing sparse records in user clicking behaviour as shown in formula 7.



190 M. Chen

SS(u) =
2

n(n + 1)
Sum(

Sum(sik(u)sjk(u))k∈dataitem√
Sum((sik(u))

2)k∈dataitem
√
Sum((sjk(u))

2)k∈dataitem
)ni<=j (7)

Based on the above measurement of degree distribution and similar mode for Web news
instances, in this module, finally, administrator should use formula 8 to infer results of the Web
news instance retrieval mode.

SearchMode(u) = DR(u)SS(u) (8)

4.3 The analysis module of Web news instance seed

The inputting content of this module is analysis results of Web news instance clicking and
retrieval mode, the outputting content of this module is sets of topic contained in massive Web
news instances and Web news seed instances, the main function of this module is to infer set of
Web news seed instances according to analysis results of Web news instance clicking and retrieval
mode, and describe corresponding topic referring to Web news key information researched in
previous job.

In this process, firstly, administrator should use formula 9 to mine set of Web news seed
instances, its weight value is more than or equal to seed threshold, secondly, based on releasing
time of Web news, the Web news seed instances should be sorted in set, finally, the corresponding
topic should be described using key information of the Web news seed instance, in following
experiment, the optimal value of seed threshold will be analysed.

SeedURL(u) = ClickMode(u)SearchMode(u) (9)

4.4 The analysis module of similar Web news instance supporting topics

The inputting content of this module is set of Web news seed instances and records using
Web news behaviour for users, the outputting content of this module is set of similar Web news
instances supporting topics, the main function of this module continues to mine set of similar
Web news instances with topic described according to sets of Web news seed instances, the
corresponding topics described, the records using Web news behaviour for users that have been
analysed above.

In this process, administrator should take Web news seed instances as core, and use the
probability of first transfer from the Web news instance to itself as possibility that whether it
is able to support topic described. If the Web news seed instance is set su, then the variable
tu indicates that whether the Web news instance is able to support topic described by su, the
variable ts indicates that whether the retrieval keyword is able to support topic described by
su. If the key information of the Web news instance can support topic described by su, then
tu = 1, otherwise tu = 0, if the retrieval keyword can support topic described by su, then ts = 1,
otherwise ts = 0. For each Web news seed instance mined, in initial state, tsu is set one, then
P(tsu = 1) = 1, and the probability is set zero for any other Web news instances supporting
topic described by su, in this way, P(ts = 1) can be used to calculate the probability arriving
su to itself, which is directly linked to the search keyword with it.

P(ts = 1) = Sum(ψsuP(tu = 1))u:(s,u)∈E (10)

ψsu =
fq(s,u)

Sum(fq(s,ui))(s,ui)∈E
(11)



Model of Network Topic Detection Based on
Web Usage Behaviour Mode Analysis and Mining Technology 191

In formula 11, ψsu expresses transfer probability form the search keyword to the Web news
instance, based on this probability, the value of P(tu = 1) can be calculated by using the following
formula for all other Web news instances that are directly connected to it.

P(tu = 1) = Sum(ψusP(ts = 1))s:(s,u)∈E (12)

ψus =
fq(s,u)

Sum(fq(si,u))(si,u)∈E
(13)

In formula 13, ψus expresses transfer probability form the Web news instance to the search
keyword, when P(tu = 1) is greater than or equal to probability threshold, then the Web news
instance can be found to support topic described by su, in order to mine set of similar Web news
instance that can support topics, in the following experiment, the optimal value of probability
threshold will be analysed.

5 The design of network topic detection algorithm

Based on model design of network topic detection above, in this section, the author designs the
mining algorithms of Web news seed instances and similar Web news instances supporting topics,
in order to make sure that the topic detection has a high accuracy for Web news, in following
experiments, the precision of algorithms will be analysed, the optimal value of parameters will
be determined.

5.1 The mining algorithm of Web news seed instances

The key information of Web news have been expressed using Web news information extraction
and analysis method researched in previous job [16], [17], [18], but what topics the users concern
are still unknown in the face of massive Web news released based on social events. Therefore,
this algorithm mainly uses set of records using Web news behaviour for users and results of Web
news information extraction and analysis, and through analysing Web news instance clicking and
retrieval mode to mine set of topic contained in massive Web news instances.

5.2 The precision analysis of topic detection under different method

This experiment compares precision of Web news topic detection under Web news instance
clicking mode analysis method, Web news instance retrieval mode analysis method and the
method proposed in this paper. As shown in figure 3, the precision represents quality of Web
news topic detection using three methods, the red column expresses precision change situation of
Web news topic detection using Web news instance clicking mode analysis method that is called
DClickMode method in this chart, from its trend, it can be known that the quality of Web news
topic detection is not high only through a single analysis for Web news instance clicking mode
with increasing number of Web news instances concerned, although the precision has a certain
improvement, but the maximum can only float on the 62.8% the blue column expresses precision
change situation of Web news topic detection using Web news instance retrieval mode analysis
method that is called DSearchMode method in this chart, from its trend, it can be known that
the quality of Web news topic detection is not also high comparing with DClickMode method
only through a single analysis for Web news instance retrieval mode with increasing number of
Web news instances concerned, although the precision has also a certain improvement, but the
maximum can also only float on the 63%, the green column expresses precision change situation of
Web news topic detection using method proposed by this paper in this chart, from its trend, it can



192 M. Chen

be known that the quality of Web news topic detection has been significantly improved, because
of integrating two analysis methods of Web news instance clicking and retrieval mode, while the
quantity of Web news instances concerned is less, although the difference of precision is not big
comparing with other two methods, the distance of precision is constantly widening among other
two methods with increasing number of Web news instances concerned, the maximum can float
on the 75.2%, this experiment shows that the quality of Web news topic detection using method
proposed in this paper is higher than DClickMode and DSearchMode method.

Figure 3: The quality of Web news topic detection under different methods

5.3 The impact analysis of Web news topic detection quality from the number
of Web news instances concerned and seed threshold

Under the increasing number of Web news instances concerned, this experiment analyses
precision change situation of Web news topic detection through adjusting seed threshold. As
shown in figure 4, the precision represents quality of Web news topic detection with increasing
number of Web news instances concerned in Y axis through adjusting seed threshold in X axis,
when the threshold value is certain, from this graph, it can be known that the quality of Web
news topic detection can increase slowly until a relatively stable trend with increasing number of
Web news instances concerned, the main reason is that when the number of Web news instances
concerned is less, the data and its relationship among them are relatively simple in the process
of analysing clicking and retrieval mode, when the number of Web news instances concerned is
increasing, the link relationship will exist among data in the process of analysing clicking and
retrieval mode, which is conducive to topic detection, so that the precision of Web news topic
detection is gradually increasing until more stable, when the number of Web news instances
concerned is certain, from this graph, it can be known that the quality of Web news topic
detection shows a trend of increasing firstly and then decreasing with increasing of threshold
value, the main reason is that when the threshold value is less, a part of inaccurate or approximate
accuracy Web news topic is likely to be detected as accurate Web news topic, when threshold
value increases to a certain stable interval, only a small number of approximate accurate Web
news topic are likely to be detected as accurate Web news topic, when threshold value is increased
to a certain value, a part of accurate Web news topic may not be detected, this experiment shows
that when the number of Web news instances concerned is 160, and seed threshold is 0.75, the
quality of Web news topic detection can reach the highest value, which is close to 78.5%.

5.4 The mining algorithm of similar Web news instances supporting topics

Although the topics concerned by users have been detected using the mining algorithm of
Web news seed instances, but what Web news instances supporting these topics the users concern



Model of Network Topic Detection Based on
Web Usage Behaviour Mode Analysis and Mining Technology 193

Algorithm 1 MiningSeedTopic
Input: UserBehavior, NewsSet, Threshold, InitialTime, T;
Output: TopicURL
MiningSeedTopic(UserBehavior ub, NewsSet ns, SystemData s);
BEGIN
UserRecord u[]=ExtractRecord(ub);
double br,cr,cm,dr,ss,sm;
GroupUserRecord<u> gur[];
TopicURL tu=new TopicURL();
gur=GroupByURL(u);
for i do=0 to gur.size()-1

br=CalculateBR(gur[i]);
cr=CalculateCR(gur[i],u);
cm=br*cr;
dr=CalculateDR(gur[i],u);
ss=CalculateSS(gur[i]);
sm=dr*ss;
if c thenm*sm>=s.getThreshold()

tu.add(ns.getFT(gur[i].url),gur[i].url);
end if
if g thenetCurrentTime()-s.getInitialTime()>=s.getT()

ReSort(tu);
ReAdjust(s.getThreshold());

end if
end for
END

are still unknown in addition to Web news seed instances. Therefore, this algorithm mainly uses
sets of records using Web news behaviour for users and Web news seed instances analysed to
mine set of similar Web news instance which can support topics.

6 The experimental analysis and results

In this section, the author carries out experimental analysis and shows experimental results
in order to validate feasibility, validity and superiority of model proposed in this paper, in
this process, the author adopts experimental environment towards to event of German A320
plane crash shown as follows. The processor is dual core, the memory is 32G, the language of
computer programming design is Java, its version is Java SE Development Kit 8, the platform of
experimental design and implementation is MyEclipse 2015, the platform of experimental data
storage and management is Microsoft SQL Server 2016.

6.1 The quality impact analysis of Web news instances mined supporting
topic from the number of Web news instances concerned and probability
threshold

Under the increasing number of Web news instances concerned, this experiment analyses
precision change situation of Web news instances mined supporting topic through adjusting
probability threshold. As shown in figure 5, the precision represents quality of Web news instances



194 M. Chen

Algorithm 2 MiningSimilarTopicURL
Input: TopicURL, UserBehavior, Threshold, InitialTime, T.
Output: TopicURL
MiningSimilarURL(TopicURL tu, UserBehavior ub, SystemData s)
BEGIN
double pts,ptu;
for i do=0 to tu.size()-1

SearchWord sw1,sw2;
WebNewsURL wnu1,wnu2;
sw1=IsExist(tu[i].get("Topicurl"),ub);
while s dow1!=NULL

for j do=0 to sw1.size()-1
pts=Calculate(sw1.get(j).position,tu[i].get("Topicurl"),ub);
if p thents>=s.getThreshold()

sw2.add(sw1.get(j));
end if
ub.set(sw1.get(j).position,pts);

end for
for j do=0 to sw2.size()-1

wnu1=IsExist(sw2.get(j).position,ub);
if ( thenwnu1=IsEqual(wnu1,wnu2))!=NULL

for k do=0 to wnu1.size()-1
ptu=Calculate(wnu1.get(k).position,sw2.get(j).position,ub);
if p thentu>=s.getThreshold()

wnu2.add(wnu1.get(k));
end if
ub.set(wnu1.get(k).position,ptu);

end for
end if

end for
sw1=IsExist(wnu2,sw2,ub);

end while
tu[i].set(wnu2);

end for
if g thenetCurrentTime()-s.getInitialTime()>=s.getT()

ReSort(tu);
ReAdjust(s.getThreshold());

end if
END



Model of Network Topic Detection Based on
Web Usage Behaviour Mode Analysis and Mining Technology 195

Figure 4: The changing trend of precision with number of Web news instances concerned and
seed threshold

mined supporting topic with increasing number of Web news instances concerned in Y axis
through adjusting probability threshold in X axis, when the threshold value is certain, from
this graph, it can be known that the quality of Web news instances mined supporting topic can
increase slowly until a relatively stable trend with increasing number of Web news instances
concerned, the main reason is that when the number of Web news instances concerned is less,
the data and its relationship among them are relatively simple in the process of analysing similar
Web news instances supporting topic, when the number of Web news instances concerned is
increasing, the link relationship will exist among data in the process of analysing similar Web
news instances supporting topic, which is conducive to instance mine, so that the precision of
Web news instances mined supporting topic is gradually increasing until more stable, when the
number of Web news instances concerned is certain, from this graph, it can be known that the
quality of Web news instances mined supporting topic shows a trend of increasing firstly and then
decreasing with increasing of threshold value, the main reason is that when the threshold value
is less, a part of inaccurate or approximate accuracy Web news instances are likely to be mined,
when threshold value increases to a certain stable interval, only a small number of approximate
accurate Web news instances are likely to be mined, when threshold value is increased to a
certain value, a part of accurate Web news instances may not be mined, this experiment shows
that when the number of Web news instances concerned is 140, and probability threshold is 0.7,
the quality of Web news instances supporting topic can reach the highest value, which is close
to 75.7%.

6.2 The process analysis of detecting Web news topic

The author illustrates effectiveness of Web news topic detection method implemented in this
paper, As shown in figure 6, in this experimental webpage, firstly, users can choose the social
event occurred that is German A320 plane crash, secondly, users can choose releasing time of Web
news reporting the social event chose, thirdly, users can choose place, object or core event related



196 M. Chen

Figure 5: The changing trend of precision with number of Web news instances concerned and
probability threshold

to the social event chose, whose selection range is imported from result of Web news extraction
and analysis, fourthly, users can input keywords in textfield component again, finally, when
users click Web news search button, the experimental platform will display webpage as shown in
figure 7. In this webpage, the users can browse retrieval results of Web news instances according
to retrieval condition chose or inputted that include title, releasing time, leading content and
keyword of the Web news instances. When users are concerned about the Web news instance,
and then click its title link, the experimental platform will display corresponding webpage that
is linked source URL, at the same time, it will also record the current user name, the search
keywords submitted, URL of clicking the Web news instance, usage time and other information
in order to detect Web news topic.

As shown in figure 8, in this experimental webpage, firstly, users can choose the social event
occurred that is German A320 plane crash, secondly, users can choose the number of topic
detected that need to be displayed, finally, when users click submit button, the experimental
platform will display top N topics description that have been detected, these topics are sorted
in accordance with releasing time of corresponding Web news seed instance, in addition, it will
also show set of Web news instances supporting every topic and category information of each
topic, Web news instances are also sorted in accordance with its releasing time in set, when users
are concerned about the Web news instance supporting topic, then click on its title link, the
experimental platform will show corresponding Web news instance browsing webpage.

As shown in figure 9, in this experimental webpage, firstly, users can choose the social event
occurred that is German A320 plane crash, secondly, users can choose the number of topic
detected that need to be displayed, finally, when users click submit button, the experimental
platform will show top N topics description detected by the way of time axis, these topics are
sorted in accordance with releasing time of corresponding Web news seed instance, in addition, it
will also show set of Web news instances supporting every topic, Web news instance are also sorted
in accordance with its releasing time, when users are concerned about the Web news instance



Model of Network Topic Detection Based on
Web Usage Behaviour Mode Analysis and Mining Technology 197

supporting topic, then click on its title link, the experimental platform will show corresponding
Web news instance browsing webpage.

Figure 6: The searching webpage of Web news

Figure 7: The clicking webpage of the Web news instance

Figure 8: The webpage of Web news topic detection

Conclusion

This paper completes a research on model of network topic detection based on web usage
behaviour mode analysis and mining technology, which takes Web news as research object, takes
web usage behaviour application technology as research core and executes process of defining
detection targets, extracting valuable network information, analysing web user usage behaviour,
mining potential topics and applying topics detected from point of innovation. This result is
important and valuable for researchers in the same or related field. In the process of model
research, design and implement, this paper proposes the mining algorithm of Web news seed
instances and similar Web news instances supporting topics in order to eliminate shortcomings
existing in previous traditional method.



198 M. Chen

Figure 9: The browsing webpage of Web news topics

The experimental analysis and results of model do key contributions for feasibility, validity
and superiority of network topic detection request, improve efficiency of understanding network
information for users, enhance availability of websites, build scientifically and improve service
functions of websites, and improve business operational efficiency and clicking rate of websites.
In a word, the process of research, design and implement model of network topic detection has
certain practical application value, which establishes real and exact foundation of corpus for
continuative research and application on Web text mining direction.

Acknowledgement

This work was supported by the National Natural Science Foundation of China under Grant
Nos.71572015, the National Natural Science Foundation of China under Grant Nos.71271209,
the Project of Philosophy and Social Sciences Planning in Beijing under Grant Nos.13JGC090.

Bibliography

[1] Zhang Ji, Li Hongzhou, Gao Qigang, Wang Hai, Luo Yonglong, Detecting anomalies from
big network traffic data using an adaptive detection approach, Information Sciences, 6(3):
96-97.

[2] Pandey Suraj, Nepal Surya, Cloud Computing and Scientific Applications-Big Data, Scal-
able Analytics, and Beyond, Future Generation Computer Systems, 29(7): 1774-1775.

[3] Zhu Zhiguo, A novel method for discovering frequent changing patterns from historical web
access data, ICIC Express Letters, 8(9): 2443-2445.

[4] Nasomyont, Tamrerk, A study on the relationship between search engine optimization fac-
tors and rank on google search result page, Advanced Materials Research, 3(4): 1462-1464.

[5] Guo Yi, Chen Hao, Microblog user ranking based on PageRank and Hadoop, WIT Trans-
actions on Information and Communication Technologies, 49(1): 1083-1085.



Model of Network Topic Detection Based on
Web Usage Behaviour Mode Analysis and Mining Technology 199

[6] Zhang Hongli, Huang Shouming, Web Information Extraction Method Based on MapRe-
duce, Journal of Anhui Science and Technology University, 27(2): 72-74.

[7] Li Wen, Zheng Bangxi, Deng Wu, Research on Web Information Extraction Model Based
on XML and DOM Technologies, Journal of Dalian Jiaotong University, 34(3): 96-98.

[8] Zhang Yaming, Tang Chaosheng, Information propagation model based on the dynamics of
complex networks in mircoblogging, Journal of Computational Information Systems, 10(1):
443-445.

[9] Wu Jiagao, Zhou Fankun, Zhang Xueying, Research of the Extraction Method of Event
Properties Based on the Combining of HMM and Syntactic Analysis, Journal of Nanjing
Normal University(Natural Science Edition), 37(1): 30-32.

[10] Yang Yuzhen, Liu Peiyu, Fei Shaodong, Zhang Chenggong, A topic link detection method
based on improved information bottleneck theory, Zidonghua Xuebao/Acta Automatica
Sinica, 40(3): 471-479.

[11] Suhara, Yoshihiko, Toda, Hiroyuki, Nishioka, Shuichi, Susaki, Seiji, Automatically gener-
ated spam detection based on sentence-level topic information, WWW 2013 Companion -
Proceedings of the 22nd International Conference on World Wide Web, 1157-1160.

[12] Pang Junbiao, Jia Fei, Zhang Chunjie, Zhang Chenggong, Unsupervised Web Topic Detec-
tion Using A Ranked Clustering-Like Pattern Across Similarity Cascades, IEEE TRANS-
ACTIONS ON MULTIMEDIA, 17(6): 843-853.

[13] Dziczkowski, Grzegorz, Wegrzyn-Wolska, Katarzyna, Bougueroua, Lamine, An opinion min-
ing approach for web user identification and clients’ behaviour analysis, IEEE Computer
Society, 79-84.

[14] Karakostas, Bill, Theodoulidis, Babis, A MapReduce architecture for web site user behaviour
monitoring in real time, DATA 2013 - Proceedings of the 2nd International Conference on
Data Technologies and Applications, 45-52.

[15] Zhang Yongheng, Feng Zhang, Fei You, A New Replacement Algorithm of Web Search
Engine Cache based on User Behavior, Applied Mathematics & Information Sciences, 8(6):
3049-3054.

[16] Chen Mo, Yang Xiaoping, Research on Model of Network Information Extraction Based on
Improved Topic-Focused Web Crawler Key Technology, Tehnicki vjesnik/Technical Gazette,
23(4): 49-54.

[17] Chen Xuegang, Research and realization of E-commerce monitor system based on focused
web crawler, Information Technology Journal, 12(17): 4033-4035.

[18] Balla, Andoena, Real-time web crawler detection, 2011 18th International Conference on
Telecommunications, 428-430.

[19] Ahmadi-Abkenari, F, A clickstream-based web page significance ranking metric for web
crawlers, 2011 5th Malaysian Conference in Software Engineering, 223-225.

[20] Chen Mo, Yang Xiaoping, Liu Ting, A research on user behavior sequence analysis based on
social networking service use-case model, International Journal of u- and e- Service, Science
and Technology, 7(2): 1-4.



200 M. Chen

[21] Chen Mo, Yang Xiaoping, Sun Meng, Zhao Yun, Research on model of network information
currency evaluation based on web semantic extraction method, International Journal of
Future Generation Communication and Networking, 7(2): 103-105.

[22] Zhu Tao, Lin Yumin, Cheng Ji, Wang Xiaoling, Efficient diverse rank of hot-topics-discussion
on social network, Lecture Notes in Computer Science, 8485(1): 522-524.

[23] Lu Ran, Xue Suzhi, Ren Yuanyuan, Zhu Zhenfang, A modified approach of hot topics found
on micro-blog, Lecture Notes in Electrical Engineering, 269(1): 603-605.