An E-commerce feedback review mining for a trusted seller’s profile and classification of fake and authentic feedback comments

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056
p-ISSN: 2395-0072Volume: 04 Issue: 03 | March -2017 www.irjet.net
© 2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 335
An E-commerce feedback review mining for a trusted seller’s profile and
classification of fake and authentic feedback comments
Sruthi sathyanandan s1, 2 Dhanya Sreedharan
1Sree Buddha college of engineering, Alappuzha, India,
2Sree Buddha college of engineering, Alappuzha, India
----------------------------------------------------------------***---------------------------------------------------------------
Abstract - Nowadays before making a purchase from an E-
commerce site we firstly browse the online reviews of
products posted by the post-purchase customers. Today E-
commerce sites uses trust models based on reputation of
each sites. The trust models are computed based on the
feedback ratings on E-commerce sites. The main problem
that arises during the computation is the “all good
reputation problem”.E-commerce sites like Amazon, EBay is
highly prone to the “all good reputation problem”. Thus we
need to use trust evaluation to compute the feedback ratings
obtained from various E-commerce sites. For this we mine
each of the feedback comments based on their dimensions
and weights.So,in order to mine feedback comments an
algorithm is used.
Key Words: Aspect mining, LDA, NLP, LEXICAL-LDA
1. INTRODUCTION
Before making a purchase from an E-commerce site what we
does is we go through each of the product reviews in that
particular site. Because online feedback reviews helps us to
find the post purchase experiences of products and services.
However we know that all the reviews we find in an E-
commerce site may not be genuine. Some reviews we find in
an E-commerce site may be fake yet they may be written to
appear as authentic. However it is a very tedious task to
differentiate between authentic and fake reviews. Our main
objective behind the paper is to find or to categorize sellers
in an E-commerce site by providing each seller’s trusted
profile. In an E-commerce site we can find a huge number of
customer reviews. Also we should take into consideration
that the reputation scores for each seller’s in an E-commerce
site may be very high. Hence it will be difficult for a
customer to find a trust worthy sellers. For E-commerce
sites like E-bay and Amazon the reputation scores will be
very high. So in this paper we consider a buyer’s feedback
reviews in free text feedback comments. For this we propose
an algorithm called Comm trust in order to mine feedback
comments.
Fig 1: Architecture of an e-commerce feedback review mining
for a trusted seller’s profile
The Proposed system can recognize the authentic review
comments .The system categorizes each seller’s based on
their trust in feedback comments. The overall feedback
comments are taken in order to find the weights and
dimension ratings. Then an overall trust evaluation is made
in order to find trusted reviews. Here star ratings from the
Feedback comments
Detection of genuine
feedback comments
Dimension rating
using mining
Calculate
dimension weights
and trust
Overall trust
evaluation
User’s star
rating
Weight of
star rating
Overall trust score for seller using
feedback comments and star rating
Seller trust profile

sites are also taken for trust computation. Then a seller’s
trust profile is computed based on all these dimensions.
Here the system deals with three main modules. They are as
follows.
I. Trust evaluation on the basis of reputation
Trust evaluation is done in order to compute or examine the
overall trust in an E-commerce feedback reviews. By
computing the trust from the review we can reduce the
reputation scores to a certain extend. For the easiness of
computing trust usually sellers and buyers are referred to as
individuals in an E-commerce system. For this we need to
compute the dimension ratings from the review comments.
Then we need to cluster these review comments
represented as dimensions .Then we need to compute the
corresponding dimension trust scores. The based on the
computed dimension based trust scores each sellers are
categorized.
II.Analysis of text feedback comments
We know that the customer feedback is the main thing that
gives a business a clear view of products in an E-commerce
site. We also know that the text review comments are
sometimes noisy and thus analysing them is a tedious task.
For each text feedback comments in an E-commerce site we
give a particular weightage based on the positive, negative
and neutral feedback comments.
III.Aspect opinion extraction
Here each text feedback review will be handled dependently
and indepently.Here the nouns and noun phrases that
frequently appear in a text feedback review is identified.
Then an opinion lexicon dictionary is constructed based on
the words in the text review. Then an overall summarization
is made from the reviews.
2. LITERATURE SURVEY
2.1 Aspect rating analysis
[1] Discusses a new approach in opinionated text analysis
problem called the Latent Aspect Rating Analysis also called
as LARA.LARA aims to analyse opinions based on the topics
of aspects in order to find individual opinions on each
aspects along with the relative view on different aspects
while forming a final judgement. This paper explains a
probabilistic rating model to solve the problem in a
convenient way. LARA is actually used to get a deeper
understanding of a review. To get detailed understanding of
review, we used a novel data mining problem called Latent
Aspect Rating Analysis (LARA). With a collection of text
feedback reviews with overall ratings LARA analyses
opinions expressed in each review in a level of topical
Aspects. It is to discover each individual reviewer's latent
rating corresponding to each aspect and to the relative
importance weight on aspects when forming a final
judgment. The use of latent aspect ratings as well as the
aspect weights in a single review can be used in a wide range
of application. The latent ratings on various aspects can
support aspect based opinion summarization. Here aspect
weights can be directly used for checking reviewers' rating
behaviours along with latent ratings and aspect weights can
support personalized aspect-level ranking.
[2] Discusses predicting aspect-level based ratings instead of
considering the overall rating. This paper firstly formulate
the whole task as a multiple aspect ranking problem and then
produces a set of numerical scores. A single score is allotted
to each aspect. Then an algorithm is used to learn various
ranking models by considering the individual aspects. The
application of opinion extraction allows users to collectively
analyse user’s opinions contained in Web documents. Here
we consider an opinion as quadruple consisting of an
opinion holder, the subject to be evaluated, the attribute from
which a subject is to be evaluated including the value of the
evaluation which expresses a positive or negative evaluation.
We consider the definition as the tip for our opinion
extraction task. Here in this paper we can describe opinion
extraction Task as opinion units consisting of four elements.
The feasibility of task definition based on an efficient corpus
study. Here we can consider a huge task as two forms of
relation extraction tasks. They are of aspect-evaluation
relation extraction and aspect-of relation extraction. Later we
can propose a machine learning-based method. This method
will combine both the contextual clues and statistical clues of
the text feedback comments.
[3] Discusses a statistical model that is able to discover topics
included in a text. This paper considers aspect
summarization based on two problems. First problem
considered is aspect identification and the second problem is
mention extraction. The goal is to achieve a set of relevant
aspects. The statistical model used in this paper is called the
Multi-Aspect Sentiment model (MAS).The model is based on
the multi-grain LDA and sentiment predictors.Multi-grain
LDA is a modified version of LDA. Here in the paper we
introduced a model of text feedback comments together with
aspect ratings in order to extract text that is to be displayed
in sentiment judgement. The model introduced will use an

aspect ratings in order to discover the suitable topics that
can extract small pieces of text explaining the aspects
without the need of any annotation in the data. Later we can
find that the model also finds the actual coherent topics and
gains an accuracy in sentences to a standard classified
model. The main area of work is to use a model for
sentiment summarization system to evaluate it at a suitable
level.[4] Discusses the problem in generation of a rated
summary which gives a view of the total ratings for major
aspects in order to gain a different perspective to the target
entity. Here the problem in generation of a rated aspect
summary is decomposed into three steps. The idea is to gain
a different prospective to the target entity is quite good
because each user’s may have different needs. However
rated aspect summarization can help user’s to make a good
decision by getting more detailed information.
[5] Discussed rating analysis without aspect keyword. Here
a unified model is introduced for LARA (Latent Aspect
Rating Analysis).This model does not need a pre-defined
aspect keyword. Instead it mines ratings based on the
weights assigned to each aspects by the reviewer. Here in
this paper we introduced a unified Latent Aspect Rating
Analysis Model (LARAM) .The model is able to explore all
text data contained in a text data. It then discovers the
overall ratings simultaneously. The model uses various
factors like topical aspects, latent based ratings
corresponding to each aspect and latent weights allotted to
on different aspects by a online feedback reviewer. Thus a
generative model for the review text and the whole rating
enables LARA task to be used in various forms of review
data without any aspect Keywords. LARAM model
introduced in this paper can efficiently solve any problem of
LARA, including identifying useful topical aspects, finding
interesting differences in overall aspect ratings in a review,
and for modelling preferences of user with relative view on
different aspects. LARAM technique can support multiple
application tasks, which includes aspect opinion
summarizing, ranking entities based on overall aspect rating
plus the analysis of user’s rating approach.
[6]Discusses modelling of the online reviews using multi-
grain topic modelling. These models works based on LDA
and PLSA.Here we introduce a framework to extract aspect
ratings from online user reviews. The multi-grain LDA
introduced by the paper considers both the local topics and
the global topics. The aim of the system is to use a method
for extracting aspects that can be rated from the overall
reviews without the help of any humans. Therefore we used
generative models of various documents, which represent
the whole document to a mixtures of latent topics. Here we
mostly consider the application of standard methods that can
be used for unsupervised modelling of documents.
Probabilistic Latent Semantic Analysis, PLSA along with the
Latent Dirichelet Allocation. The analysis allows us to find
certain limitations of models. Thus we propose a new model
called the Multi-grain LDA.
[7] Discusses jointly modelling all opinions and aspects using
MAXENT-LDA hybrid model. Here a new topic modelling
method that can independently separate opinions words and
aspects is introduced. Bag of words representation is used to
separate. MAXENT allows to us pos tags to separate words
more easily. Here we uses a topic modelling form that is able
to identify aspect as well as opinions by using a model named
Maxent-Lda hybrid model. By using a supervised model we
can find syntactic features in order to separate aspect and
opinions. The model is then evaluated on feedback review
datasets from various domains. The model introduced was
found to be good in identifying the meaningful side of aspects
compared to the previous models used. Also the model could
perform fair with even a small amount of training data set or
with any with training data set from various domain. The
model joints both aspects and opinion words.
2.2 Trust relationships from online feedback
comments
[8] Discusses maintaining trust relationships from feedback
comments. The system introduced in the paper is capable in
extracting useful negative information from millions of
feedback comments. It uses both personalises and feature
based trust evaluation methods. The core algorithm used in
the paper uses techniques from the field of natural language
processing (NLP).The algorithm introduced in the paper is
called the auction rules algorithm. It is a classification
algorithm that classify comments to negative and positive
comments based on a particular threshold value.
[9] Introduces a formalised method for trust which thus
provides a tool for detailed description. The method adopted
can be used as an artificial agent which enables the agent to
make trusted decisions. Here in this paper we used the
foundations that are needed for understanding of the trust in
a web based social networks in order to use different trusted
networks. Then two algorithms are used in calculating the
recommendations on how a person can trust another person
for a personal view in a trusted network. [10] Discusses
about making trust relationships in web-based social
networking sites. This paper explains a method for trust
computation. For that two algorithms are introduced to
produce a calculated trust based value which are highly

accurate. Then an application called trust mail is used as
variations on these algorithms in order to score messages in
user’s inbox based on the ratings in a trusted network.
.[11] Discusses the trusting multi agent systems. This paper
checks the role of each multi agent systems and then takes
into account. There are two main approaches to trust used
in this paper. First approach used is to allow all agents to
trust among themselves. The second approach used is in the
design of protocols and various interaction mechanisms.
Here the paper uses a method to decrease the impact of
malicious sites. Firstly the system calculates a trust value for
calculating the Eigen vector of matrix to get a fully
normalized trust values. For this the entire system’s history
value is taken. All the computations are done in a distributed
manner.
[12] Discusses an Eigen trust algorithm for reputation
management of the E-commerce sites. This paper introduces
a method to compute trust values globally. We use
normalization to aggregate all global trust values. The
system explained here completely relies on a centralised
system in order to manage and store all the user ratings.
This paper is completely implemented based on Eigen trust
algorithm. This algorithm explains how to normalise values.
[13] Discusses the design of feedback analysing system to
find trustworthy e-commerce seller. This paper uses a
feedback analysis system in order to analyse the feedback
comments and helps buyer’s to find out which organisation
is good for making a purchase. For this both opinion mining
and natural language processing techniques are used. For
this two aspects are used. Both positive and negative
aspects.[14] Discusses a general model for trust
representation and aggregation. This paper presents a trust
model, RLM for reputation prediction. Here the trust score is
calculated based on the positive score and the negative
score.
[15] Discusses identifying honest recommenders in e-
commerce sites. This paper uses mechanisms to filter out
untrustworthy sellers. Hence the system uses only
recommenders that contribute positively to the computed
reputation score. This paper uses a honesty checking
processing by using another dimension that may affect the
dishonest recommenders.
Acknowledgements
We are thankful to our project guide Prof Dhanya
Sreedharan and PG Coordinator Prof. Minu Lalitha Madhavu
for her remarks, suggestions and for providing all the vital
facilities like providing the Internet access and important
books, which were essential. We are also thankful to all the
staff members of the Department of Computer Science &
Engineering of Sree Buddha College of Engineering,
Alappuzha.
REFERENCES
[1]Latent aspect rating analysis on review text data: A rating
regression approach Hongning Wang, Yue lu, Chengxiang
Zhai department of computer science university of Illinois .
[2] Multiple aspect ranking using the good grief algorithm in
snyder and regina barzilay computer science and artificial
intelligent{bsnyder,regina}@csail.mit.edu
[3] A joint model of text and aspect ratings for sentiment
summarization ivan titov department of computer science
university of illinois at urbana-champaign urbana, il 61801
titov@uiuc.edu
[4] Rated aspect summarization of short comments yue lu
department of computer science university of Illinois at
Urbana department of computer science university of Illinois
at Urbana Champaign Urbana, il 61801 czhai@cs.uiuc.edu
[5] Latent aspect rating analysis without aspect keyword
supervision Hongning Wang, Yue lug, Chengxiang Zhai
department of computer science university of Illinois at
urbana-champaign Urbana il, 61801 usa {wang296, yuelu2,
chai}@cs.uiuc.edu
[6] Modelling online reviews with multi-grain topic models
Ivan titan department of computer science university of
Illinois at urbana-champaign Urbana, il 61801
titov@uiuc.edu Ryan McDonald google inc. 76 ninth avenue
new York, NY 10011 ryanmcd@google.com
[7] Jointly modelling aspects and opinions with a maxent-lda
hybrid Wayne xin zhao†, jing jiang‡, hongfei yan†, xiaoming
li† †school of electronics engineering and computer science,
peking university, china
[8] Extracting and visualizing trust relationships from online
auction feedback comments john o’donovan and barry smyth
adaptive information cluster school of computer science and
informatics university.

[9] Formalising trust as a computational concept stephen
paul marsh department of computing science and
mathematics university of stirling
[10] Inferring trust relationships in web-based social
networks jennifer golbeck, james hendler university of
maryland, college park computer science department
college park, md 20742 {golbeck, handler}@cs.umd.edu
[11] Trust in multi-agent systems school of electronics and
computer science, university of Southampton, Southamuk
esdr01r@ecs.soton.ac.utdh02r@ecs.soton.ac.uk;
nrj@ecs.soton.ac.uk
[12] The Eigen trust algorithm for reputation management
in p2p networks spender d. kava Stanford university.
[13] Design of feedback analysis system for deciding
trustworthy e-commerce organization sonali rakhi d.wajgi
m. tech student, dept. of computer science & engg.
[14] Rlm: a general model for trust representation and
aggregation ieee, ling liu, senior member, ieee and jinshu su,
member, ieee
[15] Identifying honest recommenders in reputation
systems farag azzedin king fahd university of petroleum and
minerals, information and computer science department,
dhahran, 31261, saudi arabia.

An E-commerce feedback review mining for a trusted seller’s profile and classification of fake and authentic feedback comments

More Related Content

Similar to An E-commerce feedback review mining for a trusted seller’s profile and classification of fake and authentic feedback comments (20)

More from IRJET Journal (20)

Recently uploaded (20)

An E-commerce feedback review mining for a trusted seller’s profile and classification of fake and authentic feedback comments