Fake Reviews Detection using Supervised Machine Learning

1 (IJACSA) International Journal of Advanced Computer Science and Applications,Vol. 12, No. 1, 2021 Fake Reviews Detection using Supervised MachineLearningAhmed M. Elmogy1, Usman Tariq2, Atef Ibrahim4 College of Computer Engineering and SciencesPrince Sattam Bin Abdulaziz University, KSA1,2,4 Faculty of Eng.,Tanta Universiy,Egypt1 Ammar Mohammed3 Department of Computer ScienceMisr International University, EgyptFaculty of Graduate Studies of Statistical ResearchCairo University, EgyptAbstract With the continuous evolve of E-commerce systems,online Reviews are mainly considered as a crucial factor forbuilding and maintaining a good reputation. Moreover, theyhave an effective role in the decision making process for endusers.

2 Usually, a positive review for a target object attractsmore customers and lead to high increase in sales. Nowadays,deceptive or fake Reviews are deliberately written to build virtualreputation and attracting potential customers. Thus, identifyingfake Reviews is a vivid and ongoing research area. Identifyingfake Reviews depends not only on the key features of the reviewsbut also on the behaviors of the reviewers. This paper proposes amachine Learning approach to identify fake Reviews . In additionto the features extraction process of the Reviews , this paperapplies several features engineering to extract various behaviorsof the reviewers. The paper compares the performance of severalexperiments done on a real Yelp dataset of restaurants reviewswith and without features extracted from users behaviors.

3 Inboth cases, we compare the performance of several classifiers;KNN, Naive Bayes (NB), SVM, Logistic Regression and Randomforest. Also, different language models of n-gram in particularbi-gram and tri-gram are taken into considerations during theevaluations. The results reveal that KNN(K=7) outperforms therest of classifiers in terms of f-score achieving best The results show that the f-score has increased by taking the extracted reviewers behavioral features Fake Reviews Detection ; data mining; supervisedmachine Learning ; feature , when customers want to draw a decision aboutservices or products, Reviews become the main source of theirinformation.

4 For example, when customers take the initiationto book a hotel, they read the Reviews on the opinions of othercustomers on the hotel services. Depending on the feedbackof the Reviews , they decide to book room or not. If theycame to a positive feedback from the Reviews , they probablyproceed to book the room. Thus, historical Reviews becamevery credible sources of information to most people in severalonline services. Since, Reviews are considered forms of sharingauthentic feedback about positive or negative services, anyattempt to manipulate those Reviews by writing misleading orinauthentic content is considered as deceptive action and suchreviews are labeled as fake [1].

5 Such case leads us to thinkwhat if not all the written Reviews are honest or credible. Whatif some of these Reviews are fake. Thus, detecting fake reviewhas become and still in the state of active and required researcharea [2]. Machine Learning techniques can provide a big contributionto detect fake Reviews of web contents. Generally, web miningtechniques [3] find and extract useful information using severalmachine Learning algorithms. One of the web mining tasks iscontent mining. A traditional example of content mining isopinion mining [4] which is concerned of finding the sentimentof text (positive or negative) by Machine Learning where aclassifier is trained to analyze the features of the reviewstogether with the sentiments.

6 Usually, fake Reviews detectiondepends not only on the category of Reviews but also on certainfeatures that are not directly connected to the content. Buildingfeatures of Reviews normally involves text and natural languageprocessing NLP. However, fake Reviews may require buildingother features linked to the reviewer himself like for examplereview time/date or his writing styles. Thus the successfulfake Reviews Detection lies on the construction of meaningfulfeatures extraction of the this end, this paper applies several Machine learningclassifiers to identify fake Reviews based on the content ofthe Reviews as well as several extracted features from thereviewers. We apply the classifiers on real corpus of reviewstaken from Yelp [5].

7 Besides the normal natural languageprocessing on the corpus to extract and feed the features ofthe Reviews to the classifiers, the paper also applies severalfeatures engineering on the corpus to extract various behaviorsof the reviewers. The paper compares the impact of extractedfeatures of the reviewers if they are taken into considerationwithin the classifiers. The papers compares the results inthe absence and the presence of the extracted features intwo different language models namely TF-IDF with bi-gramsand TF-IDF with tri-grams. The results indicates that theengineered features increase the performance of fake reviewsdetection rest of this paper is organized as follows: SectionII Summarizes the state of art in detecting fake III introduces a background about the Machine learningtechniques.

8 Section IV presents the details of the proposedapproach. Conclusions and future work are introduced fake Reviews Detection problem has been tackled since2007 [6]. Two main categories of features have been |P a g e(IJACSA) International Journal of Advanced Computer Science and Applications,Vol. 12, No. 1, 2021in the Fake Reviews Detection research; textual and behavioralfeatures. Textual features refer to the verbal characteristicof review activity. In other words, textual features dependmainly on the content of the Reviews . Behavioral featuresrefer to the nonverbal characteristics of the Reviews . Theydepend mainly on the behaviors of the reviewers such aswriting style, emotional expressions, and the frequent times thereviewers write the Reviews .

9 Although tackling textual featuresis challenging and crucial, behavioral features are also veryimportant and cannot be ignored as they have a high impact onthe performance of the fake review Detection process. Textualfeatures have extensively been seen in several fake reviewsdetection research papers. In [7], the authors used supervisedmachine Learning approaches for fake Reviews Detection . Fiveclassifiers are used which are SVM, Naive-bayes, KNN, k-starand decision tree. Simulation experiments have been done onthree versions of labeled movie Reviews dataset [8] consistingof 1400, 2000, and 10662 movie Reviews respectively. Also,in [9], the authors used Naive Bayes, Decision tree, SVM,Random forest and Maximum entropy classifiers in detectingfake Reviews on the dataset that they have collected.

10 Thecollected dataset is around 10,000 negative tweets related toSamsung products and their services. In [10], the authors usedboth SVM and Naive base classifiers. The authors workedon yield dataset which consists of 1600 Reviews collectedfrom 20 popular hotels in Chicago. In [11], the authors usedthe neural and discrete models with Average, CNN, RNN,GRNN, Average GRNN and Bi-directional Average GRNN deep Learning classifiers to detect deceptive opinion used dataset from [12] which contains truthful anddeceptive Reviews in three domains; namely hotels, restaurantsand doctors. All the above research works have only consideredthe textual features without any effort towards the articles have considered behavioral features in thefake Reviews Detection process.