Abstract—Inan e-commerce environment where millions of transactions take place between theproviders and users, a need for establishment of validity of the serviceprovided arises.
A customer feedback system has been provided by themarketplace operators in order to fulfill such need. But the feedback generatedmay not be always relied upon. The feedback may positively or negatively affectits sales, instead of showcasing the actual genuineness of the product orservice, in customer’s point of view. Our work proposes an enhancement totraditional feedback system by introducing an Trust Reputation System (TRS)which helps filtering out the valid customers using a set of algorithms,thereby creating a trust degree for the user.Keywords- component; Trust ReputationSystem, Opinion Mining, Sentiment Analysis.
I. INTRODUCTIONThe consumers in the online marketface the problem of filtering out the best products from a list of variety ofoptions.There are various marketplace operators who provide feedback system tohelp customer identify quality products, by reviewing the customer opinion andaccordingly choose the product. Most of the consumers buy products based onproduct reviews.This either negative or positively affect the sale of the products. Also,this paves a way for spammers fordecreasing the sale of the product . To eliminate this, the paper focuses onenhancing the feedback system by introducing the concept of trustworthiness.This can be done through Trust Reputation System.
TRS are programs that allow users to rate each other. Using such methodscan help decrease the number ofspammers, thereby potentially increasing the amount of genuine reviews. Theadvantage of such reviews is it helpsdetermining the genuiness of the product.
II. RELATED WORK Sentiment analysis has been studied in wide area of domainsuch as movie review, teaching review, product review, e-learning, hotel reviewand many more. Most scholars focused to quantitative data analysis. However,some studies have been done on qualitative data using sentiment analysis, wefound six works that mentioned the idea of using opinion mining and sentimentanalysis in education.Algorithms such as Naive Bayes, k-means and Support VectorMachine are used in opinion classification.
The paper also focuses on the truthreputation system. There exists several truth reputation system architectureshaving different algorithms to calculate the reputation score related to theproduct.Many authors 1,2,3,4,5,6 have proposed in theirwork several TRS architectures with different algorithms to calculatereputation score related to the product. Also, a few academic work on Truthreputation system has been devoted to the inclusion of the semantic analysis offeedbacks in the calculation of the trust score of the product and speciallythe trust degree of the user. Even in studies attempting to provide morecomplex reputation methods, some issues are still not taken into consideration,such as the credibility of referees, the update of the trust degree of the userat any intervention, the age of the rating and the feedback or the concordancebetween the given rating which is a scalar value and the textual feedbackassociated to it. In contrast to the mentioned TRS, our proposed designovercomes these issues and makes use of an algorithm which includes analysis of textual feedbacks in order tocalculate the trust degree of the user giving the feedback and a trustful reputation score for the product. III. OVERVIEW The consumers in the online marketface the problem of filtering out the best reviews or feedback for purchase ofthe products.
We try to eliminate the problem by listing out the best reviewsso that it becomes easy for the customers to decide on a product by analysing otherconsumer experiences, by allowing them to post their reviews. Consumers dealingwith the online market might sometimes buy substandard products. Though thee-commerce company provide facilities like return and exchange of products, theprocess becomes a tedious task sometimes. The project aims to provide thecostumers an opportunity to select the desired products based on the rating ofthe item they wish or plan on to buy, which has been evaluated on the basis ofrating and reviews contributed by the consumers with the help of a TruthReputation System (TRS).IV. FEATURES OF THE PROJECTThe Opinion Mining of our project will be based on Sentimentanalysis algorithms & methods and also on Truth Reputation System algorithm.Trust Reputation Systems (TRS) will provide the necessary information tosupport relying parties in taking the right decision in any electronictransaction.
In fact, as security providers in e-services, TRS have tofaithfully calculate the most trustworthy score for a targeted product orservice. Thus, TRS must rely on a robust architecture and suitable algorithmsthat are able to select, store, generate and classify scores and feedbacks. V.
PROPOSED WORK In the proposed architecture, for each user who wants toleave a rating (appreciation) and a feedback (semantic review), we analyse thecustomers attitude towards a number of short and selected feedbacks and storedby product in the knowledge base. This user’s review is going to be reached byany other user. Then, we suppose that we have a path relaying all the users(the nodes). As a result, we need to know the trust degree of the user anddetermine the trust degree of the feedback.4Trust Reputation System DesignA.
AlgorithmDescriptionThe customer starts by giving a rating and a textualfeedback about a specific product. When they click on submit, in order tovalidate the given information, we are going to redirect the user to anotherinterface showing this message for example: “please give us your opinion aboutthe following feedbacks before validating the information you gave below:” Inthis interface we will find chosen feedbacks from the database from differenttypes. Those feedbacks can be fabricated in order to summarize numerous usersfeedbacks stored in the database. The generated feedbacks can be stored inanother knowledge base.
So as much as we add feedbacks in the ordinarydatabase, we will fill the knowledge database with prefabricated feedbacksusing text mining algorithms and tools. However, some users can give alreadysummarized feedbacks that can directly be included in the knowledge database.Indeed, there are many text mining and data mining algorithms and tools thatcould search the most appropriate feedbacks that are first of all related tothe product and that can recapitulate and summarize most of each type of theusers? feedbacks.Actually, before sending the customers feedback andappreciation about the product to the trust reputation system, we have toverify the concordance between them in order to avoid and eliminatecontradiction or malicious programs attacking our system. In the redirectedinterface, we will display several feedbacks from different types. However, theuser can specify the number of feedbacks to be liked or disliked. Of course, wecan also specify the minimum and the maximum number of feedbacks to bedisplayed by the user.In fact, we are trying through this redirection to detectand analyse the user intention behind his intervention on the e- commerceapplication.
Hence, we examine and evaluate his intention using other prefabricated feedbacks with different types. Of course, we have already thetrustworthiness of each feedback. Consequently, we use our reputation algorithmstudied in section 4.2 in order to generate the user trust degree which playsthe role of a coefficient and then rectify his appreciation according to histrust degree and generates the score of the feedback. Indeed, each feedback hastrustworthiness in a threshold -5,5.
The closest is the trustworthiness to 5,the most trustworthy the feedback is. The closest is the trustworthiness to -5,the very untrustworthy is the feedback. If the feedback is trustworthy itsscore would be included in 0,5 else it would be included in -5,0.4 B.
TRS algorithm Reputation algorithm used in this TRS is using semanticfeedbacks analysis in order to generate a trustful reputation score for theproduct.Actually, we have 3 types of feedbacks:** Positive feedbacks: represent opinions that expressing apositive point of view about the product. Those ameliorative opinions contain apositive content concerning the product. Then, the adjective positive isreferring to the nature of the content of the feedbacks not itstrustworthiness. However, each feedback whatever is its type can have either apositive trustworthiness or a negative trustworthiness.
Either positivetrustworthiness or negative one, it is gradual: it has degrees as float in athreshold of -5.5.**Negative feedbacks: represent opinions talking negativelyabout the product. Logically, the users giving such opinions are not satisfiedof the commented product. This feedback could be telling the truth or a partfrom the truth or could be far from the truth. That’s why, each feedback hasits trustworthiness represented by a float number between -5 and5.
**mitigated feedbacks: represent feedbacks that are talkingpositively about some aspects of the product and negatively about other aspects.They are also characterised by trustworthiness included in -5.5.**contradictious feedbacks: represent feedbacks with acontradictious content for example a feedback where the user is not talkingabout the specified product but another one or he/she is affirming that thecamera of a mobile phone is great and later in the same opinion is saying thatthe camera is very bad.
In fact, we have to start by detecting thecontradictious feedbacks. Then we are in need of a semantic analysis algorithmand tool that can detect the contradiction in a specific content related to aproduct. We can personalize the analysis according to the product. Forinstance, if the user says that “the swimming pool of the hotel which doesnot afford one is not clean”, thealgorithm must be able to detect this great contradiction. We can give to thealgorithm for each product as an input the property of the algorithm; if thereis no similarity we can consider it as a contradiction.
But the agreementincludes the meaning of course. Because if the customer writes that thenegative thing about this hotel is that there is no swimming pool. He is telling the truth then obviously the presenceof an absent property in a feedback doesn’t mean that there is a contradiction.Actually, before sending the customers feedback and appreciation about theproduct to the trust reputation system, we have to verify the concordance andthe alliance between them so we don’t have contradiction.After verifying the concordance between the appreciation andthe textual feedback we are going to redirect the user to the selection ofprefabricated feedbacks. Then the user is going to click on like or dislikeaccording to each feedback. The event of click will be managed in order to getsome information needed in the calculus of the trust degree of the user.
Thefunction uses as a parameter the id of the feedback in order to get fromKnowledge base itstrustworthiness. We need to get also the previous trustdegree of the user if he has been already engaged in a transaction or he hasused the application for rating purpose. The user choices either “like” or”dislike” is an important parameter to determine his trustworthiness.4. VI. RESULT Intially, the user gives a rating and a textual feedbackfeedback about the purchased product.Then we validate the information provided through an interface.
In fact, in thisinterface we will find chosen feedbacks from the database from different types.The feedbacks can be used to summarize numerous users feedbacks stored in thedatabase. The generated feedbacks can be stored in other knowledge base. So asmuch as we add feedbacks in the ordinary database, we will fill the knowledgedatabase with prefabricated feedbacks using text mining algorithms and tools.However, some users can give already summarized feedbacks that can directly beincluded in the knowledge base.
Actually, before sending the user?s feedback andappreciation about the product to the trust reputation system, we have toverify the concordance and the alliance between them so we don’t havecontradiction. Test for measuring the contardictionin the feedback. Pseudo-code to verify the concordance between the rating and the textual feedback: Boolean concordance; concordance =Test_ concordance (int appreciation, string feedback) ; If (concordance) URL (url_feedbacks_interface); //redirection to the feedbacks interface Else URL (url_page); // we thank the user for his intervention and we put him temporally in a //blacklist for unconformity After measuring the concordance the feedback is sent toTrust Reputation System for further processing.At the final stage we get onlyfiltered feedback.Hence only genuine feedback about the product are generated. VII. CONCLUSIONLack of information regardingparticular products leads to wrong selection of product which in turn leadsto huge holes in pockets of thecustomers.
Thus we aim to provide the accurate and true reviews about theparticular products which will help customers in picking up the rightproduct. We attempt to calculate thetrust degree of the user according to his subjective choice either “like” or”dislike” and according to the feedback.Those results such as trust weight andscores help users making a decision about purchasing or not a product from ane-commerce application. However those scores are not always truthful. Then,they can falsify the weight and the ratings.
Semantic feedbacks are moremeaningful than single scores. VIII.FUTURE SCOPEThe consumers dealing with our website would be able to access precise data andreviews of the consumers feedback and use it intelligently for product selection and for buying of it aswell . This software would be useful for any similar e-commerce businessdealing with problems regarding the issues of trustworthiness of reviews.Theprovision of visual representation can be used by customers to buy genuineproducts. On some extent it would also help the marketplace operators andvendors to filter out their potential customers. In today’s time data is saidto be the biggest asset for any company or organization.
Thus, it is of immenseimportance to analyses the data and get some results out of it.IX. ACKNOWLEDGEMENTWe sincerely thank to our guide Mrs.Purvi Sankhe, our HOD Dr. Rajesh S. Bansode, our Dean Dr. Kamal Shah and ourprinciple Dr. B.
K. Mishra for his/her guidance and support for carrying outour project work.XI.REFERENCES 1 The Analysis and Prediction ofCustomer Review Rating Using OpinionMining Wararat Songpan Department of Computer Science, Faculty of Science, KhonKaen UniversityKhon Kaen, Thailand.2 A. Jøsang R. Hayward Simon Pope:Trust Network Analysis with Subjective Logic. Proceedings of the SecondInternational Conference on Emerging Security Information, Systems andTechnologies (SECURWARE 2008), Cap Esterel, France, August 2008.
Mandalay, Myanmar3 Fereshteh Ghazizadeh Ehsaei, Ab.Razak Che Hussin: Acceptance of Feedbacks in Reputation Systems: The Role ofOnline Social Interactions Information Management and Business Review Vol. 4,No. 7, pp. 391-401, July 2012 (ISSN 2220-3796).4 A New Reputation Algorithm forEvaluating Trustworthiness in E-Commerce Context Hasnae RAHIMI1, Hanan ELBAKKALI2 Information Security Research Team (ISeRT) Université Mohammed V-Souissi, ENSIAS5Co-Extracting Opinion Targets andOpinionWords from Online Reviews Based on the WordAlignment ModelKang Liu, Liheng Xu, and Jun Zhao6 A.
Gutowska and A. Sloane:Modelling the B2C Marketplace: Evaluation of a Reputation Metric fore-commerce. Proceedings of Web Information Systems and Technologies – WEBIST ,pp. 212-226, 2009.