IZVESTIYA 2016

Vol. 60
Journal of Varna University of Economics 4

AN APPROACH FOR AUTOMATIC ANALYSIS OF ONLINE STORE

PRODUCT AND SERVICES REVIEWS

Snezhana SULOVA1

JEL C880 Abstract

One of the advantages of e-commerce systems is that they enable
customers and merchants to become acquainted with product and
Keywords: services reviews. Currently in the most popular online stores there
data mining, web are hundreds and even thousands of reviews for certain goods, which
mining, opinion contain valuable information about the quality of the offered
mining, sentiment assortment. This is the reason to look for ways for their computer
analysis, Support processing. The article proposes an approach for automated analysis
Vector Machines, of customer reviews, based on natural language processing
Naive Bayes, e- technology and application of methods of machine learning.
commerce, model for analysis and its implementation with the software product
RapidMiner. RapidMiner are proposed.

Introduction
In recent years Internet was established as one of the richest and most easily ac-
cessible sources of information. The global network has a large amount of documents,
data, audio and video files, many recorded customer reviews. All these resources are
carriers of knowledge about business and after appropriate computer processing they
can contribute more detailed analyses and help to identify and explore new relation-
ships.
In the sphere of e-commerce, core business activities are carried out through dy-
namic online systems. One of the main challenges for this type of business is making
fast and accurate decisions in accordance with the changes in the market environment.
The e-commerce systems generate detailed and varied reports which are based mostly
on statistical processing of the data stored in the database. Lately for more detailed
and in-depth analysis in this area there have been used intelligent business analysis
based on both structured and unstructured data.
Practice has proven that nowadays new customers of online stores largely rely
on the opinions posted from existing customers. Besides, manufacturers and service
providers are also interested in analyzing customers' opinions to improve the quality

1
Department of Informatics, University of economics Varna, Bulgaria. e-mail: [email protected]
455
 Izvestiya
2016 Volume 60 4

and standards of products and services. All this requires the search for new and effec-
tive ways to transform unstructured data, such as customer opinions in detailed re-
ports and analyses.
The purpose of this article is to propose an approach for automated analysis of
online store product reviews, based on a study of existing technologies for natural
language processing.
. Theoretical foundations of computer technology
for natural language processing
The concept of natural language processing (NLP) is a broad term that can be
viewed as a synthesis between artificial intelligence and computational linguistics. It
is more than simple machine translation. It aims at the full understanding of the text,
checking syntactic and semantic validity of linguistic input, using the real world
knowledge to understand the participants goals and beliefs, and also speech acts,
conversations and discourse structure. (Kumar, 2011, p.4). Currently many research-
ers explore different aspects of intelligent text processing. In general, knowledge
discovery in unstructured data in the literature is known as text mining (TM) (Fayyad,
Piatetsky-Shapiro and Smith, 1996; Feldman, Sanger, 2007). This process is accom-
plished through the application of technology for data mining (DM) on unstructured
text data. Typical text mining tasks include text categorization, text clustering, con-
cept/entity extraction, production of granular taxonomies, sentiment analysis, docu-
ment summarization, and entity relation modeling (Pena-Ayala, 2014, p.37).
The accumulation of more and more information on the Web becomes a prereq-
uisite for extracting knowledge from Internet sources such as web pages. A new
concept of extracting knowledge from web resources - web mining (WM) is born.
Etzioni first used the term and defined it as use of data mining techniques to auto-
matically discover and extract information from World Wide Web documents and
services (e.g., on-line travel agents, job listings, electronic malls, etc.) (Etzioni,
1996, p. 65). Later the concept WM expanded and it now includes techniques for
testing and analyzing data on the usability of the web resource. (Cooley, Mobasher,
Srivastave, 1997; Markov, Larosed, 2007). Web mining is commonly divided into the
following three sub-areas (Cooley, Mobasher, Srivastave, 1997):
web content mining (WCM) extracting useful knowledge from the contents
of web documents;
web structured mining (WSM) extracting useful knowledge based on the
structure of web sites;

456
Snezhana Sulova.
An Approach For Automatic Analysis Of Online Store Product And Services Reviews

web usage mining (WUM) extracting useful knowledge from data on the
use of Internet resources.
Many researchers deal with WCM problems (Kosala, Blockeel, 2000; Navadiya,
Patel, 2012; Markov, Larosed, 2007). The researches are differentiated depending on
the specific research tasks and type of resources that are used. Automatic
classification of web pages appears in scientific publications (Materna, 2008),
grouping of documents (Markov, Larosed, 2007), detection of similarity between text
documents (Huang, 2008; Lakshmi, 2013), extracting opinions from text and
sentiment analysis (Liu 2012; Medhat, Hassan, Korashy, 2014; DAvanzo, Pilato,
2015; Patel, Prabhu and Bhowmick, 2015).
In recent years, mainly thanks to the development of web applications and social
networks, the Internet has accumulated a large amount of customer reviews, shared
impressions, feelings, emotions. This is the reason for many researchers to focus their
research on two interrelated areas such as: opinion mining (OM) and sentiment
analysis (SA) (fig. 1).

Fig. 1 Web Mining, Opinion mining and Sentiment analysis

The term opinion mining is introduced by Dave, Lawrence and Pennnock
(2003). They define it as a set of search results for a given item, generating a list of
product attributes (quality features, etc.) and aggregating opinions about each of them
(poor, mixed, good). After that for opinion mining there are more comprehensive
interpretations that involve different types of analysis and evaluation of text. Sadegh,
Ibrahim and Othman say that these are techniques to detect and extract subjective

457
 Izvestiya
2016 Volume 60 4

information in text documents. Opinion mining based on technology of natural

language processing and focusing on determining the perceptions, views, ideas on a
given subject, and automatic analysis aims to extract attributes, characteristics of the
objects and on their basis to determine whether comments are positive, negative or
neutral.
Sentiment analysis was mentioned for the first time in the works of Das, Chen
and Tong (2001). They use the term in automatic analysis and evaluation of text.
Subsequently plenty of studies discuss various aspects of the sentiment analysis and
in many of them the term is used as a synonym for opinion mining, because emotions
moods are extracted from the recorded comments (Pang, Lee, 2008; Sadegh, Ibrahim,
Othman 2012; Ankitkumar, Badre, Kinikar, 2014; Rahmath, 2014).
To analyze sentiments in a text Liu (2012) makes a comprehensive study and
divides methods depending on parts for analysis - a word, a phrase or the entire
document. Document level sentiment analysis attempts to classify the general position
in the document as positive, negative or neutral. It helps to establish the overall
polarity of the text, but through it one cannot assess the different emotions about
certain aspects of the document. Both supervised and unsupervised learning methods
can be used for the document level classification (Varghese, Jayasree, 2013). In the
sentence level sentiment analysis the polarity of each sentence is calculated using the
same methods. Many authors also work on the problems of the word or sentence level
sentiment analysis. In literature there are developments related to the grouping of
words in dictionaries and lexicons, for automatic classification of words, extraction of
keywords and identifying topics and others (Hu and Liu, 2004; Zhang, et. al. 2008;
Heerschop, et. al. 2011; Taboada, et. al 2011).
One of the most comprehensive surveys for opinion mining is made by Ravi K.
and Ravi V. (2015). They classify existing studies and methods used in them.
According to them SA solve problems related to: the classification of text to its
subjectivity; sentiment classification; review usefulness measurement; spam texts
detection; dictionaries and lexicons creation; aspect extraction. In the presented
systematization we cannot make a clear distinction between the first and second groups
so we believe it is more appropriate SA tasks to be presented as shown in Fig. 2.

458
Snezhana Sulova.
An Approach For Automatic Analysis Of Online Store Product And Services Reviews

Fig. 2. Main tasks solved with sentiment analysis

The subject areas of application of opinion mining and sentiment analysis are
defined well by Rahmat (2014). This type of analysis is mainly used:
by customers of online stores when they make a decision to purchase a product
or service;
by manufacturers to improve the quality of goods;
by marketers for research and analysis of moods and sentiments of customers;
by analytical discussion groups, forums, blogs and social medias
in detecting spam content on the web;
in detecting view points on certain topics.
II. Approach to automated analysis of online store customer reviews
As mentioned already in the aim of our study, we will focus our research on
analysis of customer reviews, recorded in online stores. The e-commerce systems are
dynamic web based applications and they enable interactivity and communication
with customers. Recorded customers opinions are one of the best measures of
whether the e-shop service is good and what shoppers like or dislike, whether

459
 Izvestiya
2016 Volume 60 4

products and services meet the descriptions and presentations, what else customers
want to discover in the online store, and what the general clients assessments are.
Opinion mining and sentiment analysis are important for traders, because they create
prerequisites for individual marketing to each customer and to implement better
service.
In this article for research and analysis of customers' reviews we suggest the use
of the methods of classification, first to distinguish the views of various
characteristics of the goods and then to evaluate the polarity of customer reviews
about them. The model that we use for analysis is shown in fig. 3.

Fig. 3. Model for analysis of online store product and services reviews
Usually the process of opinion mining from text data is an unconventional task,
because data is unstructured, its based on WCM and it is appropriate to analyze
online customer reviews following these steps:
1. Collecting and recording product reviews.
2. Text preprocessing of product reviews.
460
Snezhana Sulova.
An Approach For Automatic Analysis Of Online Store Product And Services Reviews

3. Application of methods for text processing.

4. Analysis and interpretation of results.
The collection of customer reviews is a feature that is integrated into most e-
commerce systems. In some systems the views we recorded directly in a web page in
html format, but in others they are recorded in the database. However, regardless of
the storage, reviews are unstructured data - text.
Text preprocessing is the process of preparing text for application data mining
methods such as: classification, clustering and others. Text documents usually have a
large amount of words that are not carriers of useful information and it is not
appropriate so analyze all words. Recommended preprocessing steps for customer
reviews are:
tokenization separating the full text review into a list of words;
lemmatization the process of bringing a word into its non-inflected dictionary
form;
stop word removal these are auxiliary words that carry little information
about the content of the text, such as: for, after, so, later, back, against;
text capitalization - transformation of characters into lowercase.
The methods that can be used in text sentiment analysis are well summarized by
Medhat et. al. (2014) (fig. 4).
Many of the cited surveys show that it is appropriate to use the methods of
supervised machine learning for opinion mining and sentiment analysis for online
store customer reviews, of which the most successfully applied methods for
categorization of text are: Support Vector Machines (SVM) and Naive Bayes (NB)
classifier (Singh, Husain, 2014; Verma, Kiranjyoti, 2015).
Many of the aforementioned studies show that for sentiment analysis for
customer reviews for the products and services it is appropriate to use the methods of
machine learning - supervised learning, of which the most successfully applied for
categorization of text are linear classifier Support Vector Machines (SVM) classifier
based on the algorithm Naive Bayes (NB) (Singh, Husain, 2014; Verma, Kiranjyoti,
2015).

461
 Izvestiya
2016 Volume 60 4

Fig. 4. Sentiment analysis methods (Medhat, Hassan, Korashy, 2014)

SVM is an algorithm that builds a model based on a given set of training
examples and assigns new examples into one category or another, making it a non-
probabilistic binary linear classifier. In classification based on the NB algorithm,
belonging to a class is calculated based on the formula for finding the conditional
probability of assigning the object to a class. Both methods are suitable for creating
and implementing a model for classifying customer reviews into two categories
positive and negative.
To test the automatic classification based on the SVM method we have used one
of the most popular tools for data mining and predictive analysis RapidMinier. It
provides interactive graphical user interface and tools for links analysis, analysis of
unstructured text and log files. The process of extracting knowledge is realized
through the build operators. This software provides more than 400 operators for
analyzing data, including classification, clustering, and association analysis.
The data used for testing the proposed approach are online reviews collected
from the Amazon.com website for tablets of brand Apple. Training data is a very
important element in the realization of machine learning, because validation of the
model is based on it. The created Rapid Miner model shows the classification of
customer views on polarity (Fig. 5).
462
Snezhana Sulova.
An Approach For Automatic Analysis Of Online Store Product And Services Reviews

Fig. 5. Classification customer reviews via SVM method

Based on the pattern, several posts are successfully tested and the result shows
that from the analyzed 6 sample reviews only one was rated as negative (Fig. 6).

Fig. 6. Result of an analysis of posts by RapidMiner

463
 Izvestiya
2016 Volume 60 4

In practice, this is the last stage of the analysis of reviews. It shows the results
and therefore the used software tools and the role of human interpretation of the
results are important.
Similarly, the model that is based on the NB algorithm could be built. We made
approbation and found that they obtain similar results.
Conclusion
The rapid development of social networking and sharing capabilities that
provide many of the applications running on the Internet is a prerequisite for the
generation of large collections of consumer reviews, impressions, shared feelings and
emotions. Intelligent business analyses of these customer reviews is important to the
business and therefore is subject to a research interest in the recent years. Since there
is no specific algorithm available for viable search for knowledge in a text, based on
the results of existing studies in this paper we propose an approach for analyzing the
reviews of online stores customers, through which expressed opinions can be
classified and conclusions about the quality of goods can be made.
The resulting new knowledge could help to improve the product range and
customer satisfaction, and for e-commerce companies it is essential, because sales
revenues largely depend on it. Furthermore, this kind of analysis can be used by
managers to create successful business strategies based on the resulting in-depth and
precise analyses and forecasts. Extracting new knowledge from Internet resources
could be an important competitive advantage for companies involved in e-commerce,
because in general it contributes to improving their business.
References
1. Ankitkumar, D., Badre, R., Kinikar, M. (2014) A Survey on Sentiment Analysis
and Opinion Mining. International Journal of Innovative Research in Computer
and Communication Engineering. 2 (11). p. 6633-6639.
2. Cooley, R., Mobasher, B. and Srivastava, J. (1997) Web Mining: Information and
Pattern Discovery on the World Wide Web. Proceedings of the International
Conference on Tools with Artificial Intelligence. p. 558-567
3. DAvanzo, E., Pilato, G. (2015) Mining social network users opinions to aid
buyers shopping decisions. Computers in Human Behavior. 51. p. 12841294.
4. Das, R., Chen, M. Yahoo! (2001) For Amazon: Sentiment Parsing from Small
Talk on the Web, EFA 2001 Barcelona Meetings. [Online] Available from:
http://ssrn.com/abstract=276189 or http://dx.doi.org/10.2139/ssrn.276189,
[Accessed: 10/9/2016].

464
Snezhana Sulova.
An Approach For Automatic Analysis Of Online Store Product And Services Reviews

5. Dave, K., Lawrence, S. and Pennock, D. (2003) Mining the peanut gallery:
Opinion extraction and semantic classification of product reviews. Proceedings
of WWW. p. 519528.
6. Etzioni, . (1996) The World Wide Web: quagmire or gold mine?
Communications of the ACM. 11. p. 65-68.
7. Fayyad, M., Piatetsky-Shapiro and Smyth, P. (1996) From Data Mining to
Knowledge Discovery in Databases. AI Magazine [Online] 17(3). p. 37-54.
Available from: https://www.aaai.org/ojs/index.php/aimagazine/article/view
File/1230/1131. [Accessed: 10/6/2016].
8. Feldman, R., Sanger, J. (2007) The text mining handbook. Advanced Approaches
in Analyzing Unstructured Data. Cambridge: Cambridge University Press.
9. HeerschoP B, et. al. (2011) Polarity Analysis of Texts using Discourse Structure.
Proceedings of the 20th ACM international conference on Information and
knowledge management. p. 1061-1070.
10. HU, M and Liu, B. (2004) Mining and Summarizing Customer Reviews,
Proceedings of the tenth ACM SIGKDD international conference on Knowledge
discovery and data mining, p. 168-177.
11. Huang, . (2008) Similarity Measures for Text Document Clustering,
Proceedings of the sixth New Zealand computer science research student
conference (NZCSRSC2008). [Online] pp. 49-56. Available from:
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.332.4480&rep=rep1&t
ype=pdf. [Accessed: 13/6/2016].
12. Kosala, R., Blockeel, H. (2000) Web Mining Research: a survey. ACM SIGKDD
Explorations Newsletter. [Online] 2(1). p. 1-15. Available from:
http://www.kdd.org/exploration_files/kosala.pdf. [Accessed: 11/6/2016].
13. Kumar, . (2011) Natural Language Processing. New Delhi: I. K. International
Publishing House Pvt. Ltd.
14. Lakashimi S. et. al. (2013), Analysis of Similarity Measures for Text Clustering.
International Journal of Engineering & Science Research, [Online] 3(8) pp
4627-463. Available from: http://www.ijesr.org/admin/upload_journal/journal_
Naga_lakshmi__33olaug13esr.pdf. [Accessed: 13/6/2016].
15. Liu, B. (2012) Sentiment Analysis and Opinion Mining, Morgan & Claypool
Publishers.
16. Markov, Z. and Larosed, D. (2007) Data Mining the Web Uncovering Patterns
in Web Content, Structure, and Usage. New Jersey: John Wiley & Sons.
17. Medhat, W., Hassan, A. and Korashy, H. (2014) Sentiment analysis algorithms
and applications: A survey. Ain Shams Engineering Journal. 5. p. 1093-1113.
465
 Izvestiya
2016 Volume 60 4

18. Mrna, J. Automatic Web Page Classification. Proceedings of Recent

Advances in Slavonic Natural Language Processing [Online] pp.8493.
Available from: http://www.fi.muni.cz/usr/sojka/download/raslan2008/6.pdf.
[Accessed: 11/6/2016].
19. Navadiya, D., PateL, R. (2012) Web Content Mining Techniques-A
Comprehensive Survey. International Journal of Engineering Research &
Technology [Online] 1(10). p. 1-6. Available from:http://www.ijert.org/view-
pdf/1950/web-content-mining-techniques-a-comprehensive-survey. [Accessed:
11/6/2016].
20. Pang, B., Lee, L. (2008) Opinion Mining and Sentiment Analysis, Foundations
and Trends in Information Retrieval. 2 (1-2). p. 1-135.
21. Patel, V., Prabhu, G. and Bhowmick, K. (2015) A Survey of Opinion Mining and
Sentiment Analysis. International Journal of Computer Applications. 131(1). p.
0975-8887.
22. Pena-ayala, A. (2014) Educational Data Mining. Applications and Trends.
Charm Heidelberg: Springer International Publishing.
23. Rahmath, H. (2014) Opinion Mining and Sentiment Analysis - Challenges and
Applications. International Journal of Application or Innovation in Engineering
& Management. 3 (5). p. 401-403.
24. Ravi, K., Ravi, V. (2015) A survey on opinion mining and sentiment analysis:
Tasks, approaches and applications. Knowledge-Based Systems. 89. p. 14-46.
25. Sadegh, ., Ibrahim, R. and Ali Othman, Z. (2012) Opinion Mining and
Sentiment Analysis: A Survey. International Journal of Computers &
Technology. 2 (3). p. 171-178.
26. Singh, P., Husain, M. (2014) Methodological study of opinion mining and
sentiment analysis techniques. International Journal on Soft Computing (IJSC).
5(1). p. 11-21.
27. Taboada, M. et. al. (2011) Lexicon-Based Methods for Sentiment Analysis,
Computational Linguistics. 37 (2). p. 267-307.
28. Tong, R. (2001) An operational system for detecting and tracking opinions in on-
line discussion. Proceedings of the Workshop on Operational Text Classification
(OTC).
29. Varghese, R., Jayasree, M. (2013) A survey on sentiment analysis and opinion
mining. International Journal of Research in Engineering and Technology. 2
(11). p. 312-317.

466
Snezhana Sulova.
An Approach For Automatic Analysis Of Online Store Product And Services Reviews

30. Verma, R., Kiranjyoti (2015) Opinion Mining and Analysis of the Techniques
for User Generated Content (UGC), International Journal of Advanced Research
in Computer Science and Software Engineering. 5(5). p. 438-441.
31. Zhang, C., Wang, H., Yao, L. Y., WU, D., Lia, Y., & Wang, B. (2008).
Automatic keyword extraction from documents using conditional random fields.
Journal of Computational Information Systems. 4(3). p. 1169-1180.

467