Knowledge Discovery from Log Data Analysis in a Multi-source
Search System based on Deep Cleaning
Fatma Zohra Lebib
1,2
, Hakima Mellah
2
and Abdelkrim Meziane
2
1
University of Science and Technology Houari Boumediene, USTHB, Algiers, Algeria
2
Research Center in Scientific and Technical Information, CERIST, Algiers, Algeria
{zmatouk, hmellah, ameziane }@mail.cerist.dz
Keywords: Log Files Analysis, Web Usage Mining, Multi-source Search System, Knowledge Extraction, Information
Source, User Profile.
Abstract: In a multi-source search system, understanding users’ interests and behaviour is essential to improve the
search and adapt the results according to each user profile. The interesting information characterizing the
users can be hidden in large log files, whereas it must be discovered, extracted and analyzed to build an
accurate user profile. This paper presents an approach which analyzes the log data of a multi-source search
system using the web usage mining techniques. The aim is to capture, model and analyze the behavioural
patterns and profiles of users interacting with this system. The proposed approach consists of two major
steps, the first step “pre-processing” eliminates the unwanted data from log files based on predefined
cleaning rules, and the second step “processing” extracts useful data on user’s previous queries. In addition
to the conventional cleaning process that removes irrelevant data from the log file, such as access of
multimedia files, error codes and accesses of Web robots, deep cleaning is proposed, which analyzes the
queries structure of different sources to further eliminate unwanted data. This allows to accelerate the
processing phase. The generated data can be used for personalizing user-system interaction, information
filtering and recommending appropriate sources for the needs of each user.
1 INTRODUCTION
Web servers record user activities and store them in
log files which include fields related to the user
requests. IP address, port number, state code, and
access time are commonly used fields in this kind of
log files (Mobasher et al., 2002). Different from the
past when log files were mostly neglected except for
urgent situations when system related problems were
experienced, nowadays they give important
information to system administrators and are
efficiently used for detailed investigations. In
addition to their uses for various purposes, intrusion
detection and prevention systems and digital
forensics analysis are the emerging application fields
of log files (Boeck et al., 2010).
The log file data represents the relevant
information reflecting the user’s interests, which can
be exploited in many applications, including
recommendation items, adaptation user services and
improving information retrieval. The users’
behaviour is hidden in these logs, which need to be
discovered, and analyzed (Facca and Lanzi, 2005).
Web Usage Mining (WUM) is the use of data
mining techniques to automatically discover and
extract information from web data (Patel et al.,
2011). The main goal of this technique is to discover
usage patterns trying to explain the users’ interests
(Hernández et al., 2017). The user’s interests can be
extracted in a popular way, from his/her own profile
(e.g. interests) or from his/her social behaviour (e.g.
tagging behaviour) (Astrain et al., 2010) (Li et al.
2008), his social network (e.g. friends) (Yang et al.
2015), or from log data (Carman et al. 2010).
However, detecting user’s interests is a crucial
problem (Milicevic et al., 2010). In a multi-source
retrieval system, several information sources can
contribute to provide the user with relevant results in
response to his/her query. To improve the system
performance and provide the personalized results to
each user’s profile, it is crucial to collect information
characterizing different needs and interests of users.
The main objective of this work is to discover
the user’s interests from large log data of the SNDL
1
1
National online Documentation System (www.sndl.cerist.dz)
Lebib, F., Mellah, H. and Meziane, A.
Knowledge Discovery from Log Data Analysis in a Multi-source Search System based on Deep Cleaning.
DOI: 10.5220/0008121102570264
In Proceedings of the 15th International Conference on Web Information Systems and Technologies (WEBIST 2019), pages 257-264
ISBN: 978-989-758-386-5
Copyright
c
2019 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
257
system that allows to access different national and
international electronic documentations covering all
the areas of education and scientific research. The
SNDL system covers several information sources
and databases, such as Springer Link, Science Direct
and ClinicalKey, belonging to different domains of
scientific research, including computer science,
medicine, and law. The search history of users is
recorded in log files, including the queries submitted
to sources and the documents downloaded from
these sources. These log files may contain a large
amount of raw data, which must be transformed into
a meaningful format, for example to obtain users’
preferences and behaviour, and to understand in-
depth the interests of users. An approach based on
WUM techniques is proposed to analyze the SNDL
log files and extract useful information to model an
accurate user-interests profile. The user profile can
be used further to customize the search results for
each user, to filter the information based on the
user's interests, or to recommend the appropriate
sources or documents to a specific user.
The remainder of this paper is structured as
follows. Section 2 discusses context and related
work. Section 3 presents the proposed approach.
Section 4 describes implementation of the proposed
approach and Section 5 concludes the paper.
2 RELATED WORK
A log file is a text file containing all the events
detected by a particular process, including
monitoring a network and application. The events,
including click, add to cart and execution of an
application are dated and in chronological order. It
contains all the information that has been defined as
relevant by the programmers.
Log files are a valuable source of information
that can be exploited for multiple purposes such as
issues related to computer security and systems or
applications functioning, improving the use of a site
or a web application, optimizing the performance of
a system and identifying user profiles.
However, it is impossible to manually analyze all
this data, which are too large. Web Usage Mining is
the application of data mining techniques to discover
usage patterns from Web data, in order to understand
and better serve the needs of Web-based
applications. WUM consists of three phases, namely
preprocessing, pattern discovery, and pattern
analysis (Srivastava et al., 2000). Preprocessing
consists of converting the usage, content, and
structure information contained in the various
available data sources into the data abstractions
necessary for pattern discovery. Pattern discovery
draws upon methods and algorithms developed from
several fields such as statistics, data mining,
machine learning and pattern recognition. Pattern
analysis is the last step in the overall WUM process.
The motivation behind pattern analysis is to filter
out uninteresting rules or patterns from the set found
in the pattern discovery phase.
Many researchers have widely used WUM
techniques to discover interesting and frequent user
navigation patterns from web server logs. In
(Sujatha and Punithavalli, 2012), four types of
clustering approaches are investigated in web log
files to improve the quality of clustering for user
navigation pattern in WUM systems, for predicting
user’s intuition in the large web sites. In (Vellingiri,
et al., 2015), authors proposed an approach for
discovering web user’s navigation patterns and
analyzing of web usage data using Weighted Fuzzy
Possibilistic C-Means (WFPCM) for clustering and
Adaptive Neuro-Fuzzy Inference System with
Subtractive Algorithm (ANFIS-SA) for
classification. Singh and Meenu (Singh and Meenu,
2017) presented visitor pattern analysis of
educational institution web log data using WUM for
discovering patterns and generating the reports. In
(Dwivedi, 2017), log data has been analyzed using
one of the most popular web log expert software tool
(WebLog Expert). The process of web log analysis
consists of data collection, preprocessing, pattern
discovery, pattern analysis, and result analysis
visualization. Jaya Kumar and Alagarsamy (Jaya
Kumar and Alagarsamy, 2013) discussed about the
web usage mining, different file format of log data
and carried out the analysis task with the help of
web log expert tool. In (Shanthi and Rajagopalan,
2013), an automatic discovery strategy for web
server log information using the web page collection
algorithm is proposed. The approach uses sequential
pattern mining technique to identify the sequence of
navigational patterns of web users. The extracted
patterns allow to understand the customer interests.
In (Mahoto et al., 2016), WUM has been used to
predict users’ web navigational behaviour. Two
main techniques, cluster analysis and sequential
pattern mining, have been used to get the desired
navigational patterns from real web log datasets.
Some of these approaches used existing web log
tools for attending specific objectives or generating
the statistics. There are various tools (Kaur and
Aggarwal, 2015) for analyzing log files such as
Google analytics, Stat Counter, Deep Log Analyzer,
Awstates and Web Log Expert. What can be seen is
WEBIST 2019 - 15th International Conference on Web Information Systems and Technologies
258
that some of the cited works focus on the pre-
processing step of the Web Data while others try to
concentrate on the behavioural models of Internet
users visiting websites.
The weaknesses in existing tools are summarized
in the following points:
Their configuration is not compatible with all
needs and requires some skills;
Only statistics results are provided, including
number of visitors and the amount of data
downloaded;
No knowledge on the log file content is
provided;
No structuring of data in a database is possible;
Especially, the SNDL system manages several
resources that require in-deep skills.
3 THE PROPOSED APPROACH
The proposed approach allows to extract useful
information that reflects user’s profile and interests
from the SNDL log files. The information needed to
be extracted from these log files is as follows:
User introducing the query (his/her name);
Session;
Date and time;
User query (search / download);
Keywords (query terms);
Sources accessed by a user.
The proposed approach consists of two major
steps, which are: Log file pre-processing and Log
file processing, as shown in Figure 1.
i) The first step, i.e., log file pre-processing, allows
to prepare the log file for further processing. It
consists of the following three sub-steps:
Log file cleaning;
Log event identification;
Activities segmentation by session.
ii) The second step, i.e., log file processing, allows
to perform deep analysis through the following sub
steps:
Extracting and partitioning homogenous data;
Identifying the source targeted by each query;
Keywords extraction.
Before describing the different steps of the proposed
approach, the log file structure considered in this
study is presented in the following section..
Figure 1: General schema of the proposed approach.
3.1 Structure of the SNDL Log File
A log file of the SNDL system is of a common log
format. An event represents a line in a log file.
Figure 2 shows an example of an event of this file.
Figure 2: Example of an event in the SNDL log file.
An event is composed of the following attributes:
41.98.29.193: IP address of the user;
ilWvTPtsaK6yDwC: session ID;
sahnounefoudil: user name;
(20/Feb/2013:00:00:18 +0100): date and time of
query submission;
GET : method used by the protocol;
http://www.springermaterials.com:80/docs/pdf/1
0000866_118.html : URL of the source;
HTTP/1.1: protocol used and its version;
200: status or code returned from the server;
15413: size of the requested page in bytes.
The "common" structure of an event of the
SNDL log file is given by the following form.
3.2 Log File Preprocessing
3.2.1 Log File Cleaning
Log data is typically noisy and unclear, so data
cleaning is an essential process for effective mining
process. Data cleaning allows to remove irrelevant
items stored in the log files that may not be useful
Knowledge Discovery from Log Data Analysis in a Multi-source Search System based on Deep Cleaning
259
for analysis purposes (Cooley et al., 1999; Cooley et
al., 1997), such as access of JPEG, GIF file, Java
Scripts, other audio/video files, non-human accesses
(accesses made by web robots), and accesses with
failed HTTP status codes.
Since the interesting data in the log files is the
queries submitted to the sources and the documents
downloaded from the sources and not in any system
generated data, it is important to keep only this data
in the log files after the cleaning process.
The cleaning process is divided into two
different steps, namely conventional cleaning and
deep cleaning, where:
Conventional cleaning: is similar to cleaning
methods used in most data mining techniques
which consist in removing multimedia files, style
pagefiles, java script files, error codes, and web
robot requests;
Deep cleaning: analyzes the queries structure and
eliminates those that are not relevant to our
analysis, which do not contain information on
user queries and downloaded documents.
Before describing these two cleaning steps, the
following mathematical formulations are given.
Formal Description of the Cleaning Process: Let
F be a log file in the common format. Let E be the
set of events of F. E = {e
1
, e
2
, e
3
… e
m
}, e
k
is an
event of F where 1 ≤ k ≤ m. Let
t
be an event of F
being processed. The cleaning process is performed
according to the following rule:
if (
t
satisfies condition c) then (E = E - {
t
})
This means that if the current event meets the
predefined condition then remove this event from
the log file.
Conventional Cleaning: This first cleaning step is
to eliminate lines from the log file that are generally
useless in almost every data mining process because
they contain no relevant information.
Conventional cleaning rules are defined using the
following predefined functions, namely:
Status: e
k
Status (e
k
): returns the value of the
"Status" attribute of e
k
Query: e
k
Query (e
k
): returns the value of the
"Query" attribute of e
k
Size: e
k
Size (e
k
): returns the value of the "Size"
attribute of e
k
User: e
k
User (e
k
): returns the value of the "User"
attribute of e
k
Extension: Query (e
k
): Extension ((Query (ek)):
returns the file type of the query associated with e
k
File: Query (e
k
) File (Query (e
k
)): returns the file
name of the query associated with e
k
Let D be a set of file extensions, such as: D = {jpg,
jpeg, gif, png, bmp, ico, mp3, wma, wav, ogg, mp4,
mkv, avi, WebM, CSS, js}
Examples of conventional cleaning rules are
presented in Table 1.
Table 1: Conventional cleaning rules.
Rule
number
Rule
Description
1
(Status(
) < 200
Status (
) > 399)
E = E – {
}
Remove the event with status
codes over 299 or under 200
(the failed HTTP status code)
2
Size(
) = 0 E =
E – {
}
Remove the event that does
not contain relevant
information, including pages
used to allocate a session,
open an account, and redirect
to another page
3
Extension(
) D
E = E – {
}
Remove the events have
filename extension of gif,
jpeg, css, and so on (e.g.
multimedia files)
Algorithm 1 describes the conventional cleaning
process.
Algorithm 1: Conventional cleaning.
Input: the log file F
Output: the log file after the conventional cleaning F
c
begin
while (there are still lines in F) do
= Read(F) // The current line
if (Status(
) < 200 Status (
) > 399) then E = E
– {
}
else if (Size(
) = 0) then E = E – {
}
else if (Extension(
) D) then E = E – {
}
else if (File(
) = robots.txt) then E = E – {
}
else if (User(
) = ‘-‘) then E = E – {
}
endwhile
Remove the “IP”, “Method”, “Status”, “Protocol” and
“Size” fields from F
F
c
= F
End
Deep Cleaning: is performed after the conventional
cleaning, to further eliminate lines that are not
interesting for this analysis. The events that do not
contain information on user query and documents
downloaded are removed. This can be done by
checking the query content in each event. As the NDL
system manages several sources, including Springer
Link, Science Direct, and IEEE, different query
formats can be found in the log file (each source has
its own query format). This step requires
WEBIST 2019 - 15th International Conference on Web Information Systems and Technologies
260
understanding the query structure of each SNDL
source in order to eliminate the irrelevant queries
from the log file after the conventional cleaning.
Two types of query are distinguished, namely:
Download query: used to download PDF
documents from SNDL sources;
Search query: containing the keywords entered by
the user for SNDL sources.
The study of the queries structure of SNDL
sources is based on the main composition of the
HTTP query. Indeed, an HTTP query is decomposed
in the following attributes (Shepperd et al., 2018):
Protocol used (http protocol);
Domain name (Web server or source);
Path to the file (is the path to the resource);
Parameters (Each Web server has its own rules
regarding parameters).
Figure 3 shows an example of query consisting of
the following attributes:
Protocol: http;
Domain name: link.springer.com;
Path: /search;
Parameters: facet-discipline.
Figure 3: Example of query from the SNDL log file.
Two pattern dictionaries are created, describing all the
query patterns of the SNDL sources, which are:
Search patterns dictionary: describes the search
queries (Table 2);
Download patterns dictionary: describes the
download queries (Table 3).
Table 2: An excerpt from the search patterns dictionary.
Source name
Domain
Path(s)
Relevant
parameters
ACM
dl.acm.org
/results.cfm
Query
Springer Link
link.springer.
com
/search
query; term
zbmath.org
zbMATH
/;/authors/;/journals/;
/classification/
q;f
Table 3: An excerpt from the download patterns dictionary.
Domain
Source name
PDF Marker
portalparts.acm.org
ACM
.pdf
delivery.acm.org
.pdf?
www.cairn.info
CARIN
load_pdf
ieeexplore.ieee.org
IEEE
.pdf?
Each SNDL source is represented by patterns able
to determine if the query is a search query, download
query or other.
a) Search Pattern: characterized by:
Domain of URL (the concerned source);
Source name;
List of paths to pages dedicated to search;
Search parameters containing the keywords.
b) Download Pattern: characterized by:
Domain of URL;
Source name;
Download parameters (download markers).
Deep Cleaning Process: For this step, the following
functions are defined.
Query: e
k
Query(e
k
) : returns the value of the
“Query” attribute of e
k
Domain: Query (e
k
) Domain(Query(e
k
)): returns
the value of the “Domain” attribute of the query
associated with e
k
Path: Query (e
k
) Path(Query(e
k
)): returns the value
of “Path” attribute of the query associated with e
k
Params: Query (e
k
) Params(Query(e
k
)): returns the
value of “Parameters List” attribute of the query
associated with e
k
Let D
R
be the set of domains registered in the search
patterns dictionary. Let D
T
be the set of domains
registered in the download patterns dictionary. Let M
be the set of the download markers in the download
patterns dictionary. Let N be the set of the paths in the
search patterns dictionary. Let P be the set of the
parameters in the search patterns dictionary. Table 4
describes the deep cleaning rules.
Table 4: Deep cleaning rules.
Rule
number
Rule
Description
1
(Domain(Query(
)) D
R
D
T
) E =
E – {
}
Remove the event
where the query is
not associated with
any of the SNDL
sources
2
Path(Query(
))=Ø
Params(Query(
) =
Ø)) E = E – {
}
Remove the event
where the query
contains no path or
parameter
3
(Domain(Query(
))
D
T
x M, x
Path(Query(
)))
(Domain(Query(
))
D
R
(Path(Query(
)) N
y P, y
Params(Query(
))))
E = E – {
}
Remove the event
where the query is
neither a download
query nor a search
query
Knowledge Discovery from Log Data Analysis in a Multi-source Search System based on Deep Cleaning
261
Algorithm 2 describes the deep cleaning process.
Algorithm 2: Deep cleaning.
Input: the log file F
c
Output: the log file after the deep cleaning F
n
begin
while (there are still lines in F
c
) do
= Read(F
c
) // the current line
if (rule 1 = true) then E = E – {
} ;
else if (rule 2 = true) then E = E – {
} ;
else if (rule 3 = true) then E = E – {
}
endwhile
F
n
= F
c
End
3.2.2 Log Event Identification
After removing irrelevant events from log file, this
step consists in identifying the type of each event.
An event can be either a search event or a download
event according to the type of query. Deep cleaning
not only eliminates unnecessary queries, but also
identifies search queries and download queries
according to rule 3 (Table 4).
3.2.3 Activities Segmentation by Session
It consists in segmenting user activities by session
before extracting the data to distinguish the different
visits of each user. At time t, a session identifier (ID)
is assigned to a user connected to the SNDL system.
The user will keep this identifier throughout his/her
activity, at the break of the latter, or by closing the
browser, a new session is assigned to the user. The
probability that two users have the same session ID
tends to 0. Due to the complexity of identifier
generation, it is assumed below that if the session ID
is known, then the name of the user is also known.
3.3 Log File Processing
3.3.1 Extracting and Partitioning Data
The relevant data is extracted and partitioned into
two sets, namely a set of search events and a set of
download events, such as:
A search event is characterized by:
Session ID;
User name;
Source name;
Date and time the user accessed the system;
Search query;
List of keywords entered by the user.
A download event is characterized by:
Session ID;
User name;
Date and time the user accessed the system;
Source name;
Download query.
3.3.2 Source Identification
It consists in identifying the source targeted by each
query. For this, patterns dictionaries are used to
identify the source name according to the domain in
the URL query.
3.3.3 Keywords Extraction
To extract the keywords from user’s search query,
the search patterns dictionary (Table 2) is used. The
user query may contain special characters, such as
“:”,”/”, “.”, which are encoded in URL format. For
this, a URL decoding is performed to convert these
characters to original format.
Example: Consider the following keywords entered
by the users: greenhouse%2C+heat+transfer.
The characters “+” and “% 2C” are the encoding of
the space and the comma respectively. After
decoding these special characters, the following
sentence was obtained: greenhouse, heat transfer.
4 IMPLEMENTATION
The proposed approach is implemented in a client-
server architecture (Figure 4). The client accesses
the application via a user account. A remote server
hosts database containing very large amount of data.
Figure 4: Client/Server Architecture.
Three main actors who interact with the system
are identified, namely:
User: can only download the extracted data;
Administrator: analyze the log files;
Maintainer: manages the sources and maintains
the two patterns dictionaries (add a new source).
Three log files of different sizes are analyzed.
Table 5 shows the amount of data before and after
WEBIST 2019 - 15th International Conference on Web Information Systems and Technologies
262
cleaning process, as well as the remaining data in
each file. Various others statistics can be generated,
such as the number of events detected regarding
error 404, image files, and JavaScript files.
Table 6 shows the results of the cleaning process,
illustrating the percentage of the amount of data that
was removed by conventional cleaning and deep
cleaning, as well as the remaining data in each log
file after the cleaning process. It can see that the
proposed approach eliminates a large amount of data
that is not important for the analysis which speeds
up the data extraction process. Deep cleaning
removes a considerable amount of unwanted data,
e.g. 40% from log file 1 as shown in Table 6.
Table 5: Description of the three log files before and after
the cleaning process.
Description
Log File 1
Log File 2
Log File 3
Size (MB)
62.2
140
150
Amount of data
2
before
cleaning
336283
421320
412146
Amount of data after
conventional cleaning
199268
286446
285412
Amount of data after
deep cleaning
133575
119628
122035
Total amount of
deleted data
332843
406074
407447
Amount of data
remaining
3440
15246
4699
Table 6: Statistical results of the cleaning process.
Description
Log File 1
Log File 2
Log File 3
Data deleted by
conventional cleaning
59%
68%
69%
Data deleted by deep
cleaning
40 %
28%
30%
Remaining data
1%
4%
1%
Table 7: The extracted data from in each log file.
Description
Log File 1
Log File 2
Log File 3
Research data
2350
4150
3500
Download data
1100
10150
1100
Table 7 shows the amount of data extracted from
the three files, namely research data and download
data. The extracted data consists of downloaded
documents and keywords used by the users, which
constitute relevant information that represents the
user’s interests. This data is structured as Mysql data
that can be downloaded as an Excel file or directly
access using the SQL language.
In addition to discovering the user's interests
from large log data, the proposed approach allows
the preparation of relevant statistics that provide
2
Amount of data represents the number of events.
information about SNDL users, such as the most
used keywords, their activities on SNDL sources, the
most downloaded documents and the most requested
documents (or sources). These statistics help to
improve research on the SNDL system. Table 8
shows an extract of the statistical results of the most
used keywords by SNDL users. Table 9 shows the
frequency of access to the sources, by calculating the
number of users visiting each of the SNDL sources.
Table 8: Example of most searched keywords.
Word
Number of time of search
Dirac oscillators
186
Microfluidic
135
true
90
Fault diagnisis
63
Neutron detection
60
Table 9: Sources access frequency.
Source
Number of visits
ACM
121
Aluka
8
IOP Science
103
Springer Materiels
33
IEEE
218
Science Direct
1040
5 CONCLUSION
Log files are a relevant data source that can be used
in various applications, such as customizing a search
system. In this paper, WUM techniques are used to
analyze the user’s behaviour, recorded in large log
files of a multi-source search system. The proposed
approach includes the preprocessing step that deletes
the unwanted data and the processing step that
extracts the relevant data describing the user’s
interests. In the pre-processing phase, a deep
cleaning is proposed to remove the largest amount of
irrelevant data and therefore reduce the time
required for the processing phase and get accurate
results. The extracted data is used to create a user
profile and understand the user’s activities via the
generated reports.
The sources management is considered as a
limitation of the proposed approach because it
requires human intervention to add new sources and
update the corresponding patterns dictionaries. In
future, it is planned to automate the sources
management through the use of learning methods.
Finally, the generated user data can be used later
to form clusters, to facilitate cooperation, to emerge
social networks that did not exist before, in order to
Knowledge Discovery from Log Data Analysis in a Multi-source Search System based on Deep Cleaning
263
personalize the interaction with the system or to
recommend the appropriate sources to the user.
REFERENCES
Astrain, J. J., Cordoba, A., Echarte, F., & Villadangos, J.,
2010. An algorithm for the improvement of tag-based
social interest discoery. SEMAPRO 2010 : The Fourth
International Conference on Advances in Semantic
Processing. (pp. 49-54).
Boeck, B., Huemer, D., & Min Tjoa, A., 20-23 April 2010.
Towards More Trustable Log Files for Digital
Forensics by Means of “Trusted Computing”. 24th
IEEE International Conference on Advanced
Information Networking and Applications (AINA),
(pp. 1020-1027). DOI: 10.1109/AINA.2010.26
Carman, M. J., Crestani, F., Harvey, M., & Baillie, M.,
2010. Towards query log based personalization using
topic models. In Proceedings of the 19th ACM
international conference on Information and
Knowledge Management, (pp. 1849-1852). Toronto,
ON, Canada.
Cooley, R., Mobasher, B., & Srivatsava, J., 3-8 Nov. 1997.
Web mining: Information and pattern discovery on the
World Wide Web. 9th IEEE Inernational Conference
on Tools with Artificial Intelligence. Newport Beach,
(pp. 558-567). CA, USA, USA.
Cooley, R., Mobasher, B., & Srivastava, J., 1999. Data
Preparation for Mining World Wide Web Browsing
Patterns. Knowledge and Information Systems, 1(1), 5-
32.
Dwivedi, S., 2017. User’s Activity Analysis From
Weblog: Web Log Expert. International Journal of
Engineering Sciences & Research Technology
(IJESRT), 6 (12), 547-558.
Facca, F. M., & Lanzi, P. L., 2005. Mining interesting
knowledge from Weblogs: A survey. Data Knowl.
Eng., 53(3), 225-241.
Hernández, S., Álvarez, P., JFabra, J., & Ezpeleta, J.,
2017. Analysis of Users’ Behavior in Structured e-
Commerce Websites, Access IEEE, vol. 5, pp. 11941-
11958.
Jaya Kumar, V., & Alagarsamy, K., 2013. Analyzing
server log file using web log expert in web data
mining. International Journal of Science, Environment
and Technology, ISSN 2278-3687(O) 2(5), 1008-
1016.
Kaur, N., & Aggarwal, H., 2015. A Comparative Study of
WUM tools to Analyze User Behaviours Pattern from
Web Log Data. International Journal of Advances in
Engineering Research, 10 (VI).
Li, X., Guo, L., & Eric Zhao, Y., 2008. Tag-based social
interest discovery. International Conference on World
Wide Web, (pp. 675-684), Beijing, China.
Mahoto, N. A., Memon, A., Memon, M., Teevno, M. A.,
2016. Extraction of Web Navigation Patterns by
means of Sequential Pattern Mining. Sindh Univ. Res.
Jour. (Sci. Ser.) 48 (1), 201-208.
Mobasher, B., Dai, H., Lou, T., Nakagawa, M., 2002.
Discovery and evaluation of aggregate usage profiles
for web personalization. Data Mining and Knowledge
Discovery, 6, 61-82.
Milicevic, A. K., Nanopoulos, A., Ivanovic, M., 2010.
Social tagging in recommender systems: a survey of
the state-of-the-art and possible extensions. Artificial
Intelligence Review, 33(3), 187-209.
Patel, K. B., Chauhan, J. A., Patel, J. D., 2011. Web
Mining in E-Commerce : Pattern Discovery, Issues
and Applications. International Journal of P2P
Network Trends and Technology, 1 (3), 40-45.
Shanthi, R., & Rajagopalan, S. P., 2013. An Efficient Web
Mining Algorithm To Mine Web Log Information.
International Journal of Innovative Research in
Computer and Communication Engineering, 1(7),
1491-1500.
Shepperd E. et al., 2018. Web documentation: What is a
URL?https://developer.mozilla.org/en-US/docs/Learn/
Common_questions/What_is_a_URL. Accessed 06
December, 2018.
Singh, S. P., & Meenu, 2017. User behavior Analysis from
Web Log using Web Log Expert Tools. International
Journal of Innovations & Advancement in Computer
Science IJIACS ISSN 2347 – 8616, 6 (1), 58-68.
Srivastava, J., Cooley, R., Deshpande, M., & Tan, P. N.,
2000. Web Usage Mining: Discovery and Applications
of Usage Patterns from Web Data. ACM SIGKDD
Explorations Newsletter, 1(2), 12-23.
Sujatha, V., & Punithavalli, 2012. Improved user
Navigation Pattern Prediction Technique from Web
Log Data. Procedia Engineering, 30, 92-99.
Vellingiri, J., Kaliraj, S., Satheeshkumar, S., & Parthiban,
T., 2015. A Novel Approach for User Navigation
Pattern Discovery and Analysis For Web Usage
Mining. Journal of Computer Science, 11(2), 372-382.
Yang, P., Song, Y., & Ji, Y., 2015. Tag-Based User
Interest Discovery Though Keywords Extraction in
Social Network. Y. Wang et al. (Eds.): BigCom 2015,
363-372.
WEBIST 2019 - 15th International Conference on Web Information Systems and Technologies
264
