Detection of Fake Online Reviews Using Semi-Supervised and Supervised Learning

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 08 Issue: 04 | Apr 2021 p-ISSN: 2395-0072 www.irjet.net DETECTION OF FAKE ONLINE REVIEWS USING SEMI-SUPERVISED AND SUPERVISED LEARNING N. KUMARAN1, CHAPALAMADUGU HARITHA CHOWDARY2, DEVARAPALLI SREEKAVYA3 1N. KUMARAN, Assistant professor, Dept. of Computer Science and Engineering, SCSVMV (Deemed to be University), Tamil nadu, India 2CHAPALAMADUGU HARITHA CHOWDARY, 4year (B.E), Dept. of Computer Science and Engineering, SCSVMV (Deemed to be University), Tamil nadu, India 3DEVARAPALLI SREEKAVYA, 4year (B.E), Dept. of Computer Science and Engineering, SCSVMV (Deemed to be University), Tamil nadu, India ---------------------------------------------------------------------***---------------------------------------------------------------------Abstract - Social media monitoring has been growing day by day so analyzing of social data plays an important role in knowing customer behavior. So we are analyzing Social data such as Twitter Tweets using sentiment analysis which checks the attitude of User review on movies. This paper develops a combined dictionary based on social media keywords and online review and also find hidden relationship pattern from these keyword. In recent years, shopping online is becoming more and more popular. Online reviews play a very important role in today's ecommerce decision making business. A large part of the customer population i.e. customers read reviews of products or stores before deciding where to buy and where to buy. Since writing fake reviews / frauds comes with significant gains, there has been a huge increase in fake spam views on online review websites. Poor basic reviews or fake reviews or spam review reviews are not true. A good review of the target item can attract more customers and increase sales; Poor reviews of the target item may result in lower demand and decreased sales. This false / fraudulent review was deliberately written to mislead potential customers in order to induce / deceive or defile their prominence. Our work aims to identify whether the review is false or factual. Naïve Bayes Classifier, Eristic Regression and Support Vector Machines are the classifiers used in our work. today many companies have been using Social Media Marketing to advertise their products or brands, so it becomes essential for them that they can be able to calculate the success and usefulness of each product [2]. For Constructing a Social Media Monitoring, various tool has been required which involves two components: one to evaluate how many user of their brand are attracted due to their promotion and second to find out what people thinks about the particular brand. To evaluate the opinion of the users is not as easy as it seems to all users. For evaluating their attitude may requires to perform Sentiment Analysis, which is defined as to identify the polarity of customer behavior, the subjective and the emotions of particular document or sentence. To process this we need Machine Learning and Natural Language Processing methods and this is place where most of the developers facing difficulty when they are trying to form their own tools. Over the recent years, an emerging interest has been occurred in supporting social media analysis for advertising, opinion analysis and understanding community cohesion. Social media data adapts to many of the classifications attributed for “big-data” – i.e. volume, velocity and variety. Analysis of Social media needs to be undertaken over large volumes of data in an efficient and timely manner. Analysing the media content has been centralized in social sciences, due to the key role that the social media plays in modelling public opinion. This type of analysis typically on the preliminary coding of the text being examined, a step that involves reading and annotating the text and that limits the sizes of the data that can be analysed. With the development of Web, more and more people are connecting to the Internet and becoming information producers instead of only information consumers in the past, resulting to the serious problem, information overloading. There is much personal information in online textual reviews, which plays a very important role on decision processes. For example, the customer will decide what to buy if he or she sees valuable reviews Keywords: e-commerce, product recommender, product demographic, microblogs, recurrent neural networks 1. INTRODUCTION Nowadays, Social media is becoming more and more popular since mobile devices can access social network easily from anywhere. Therefore, Social media is becoming an important topic for research in many fields. As number of people using social network are growing day by day, to communicate with their peers so that they can share their personal feeling everyday and views are created on large scale. Social Media Monitoring or tracking is most important topic in today’s current scenario. In © 2021, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 650 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 08 Issue: 04 | Apr 2021 p-ISSN: 2395-0072 www.irjet.net posted by others, especially user‟s trusted friend. People believe reviews and reviewers will do help to the rating prediction based on the idea that high-star ratings may greatly be attached with good reviews. Hence, how to mine reviews and the relation between reviewers in social networks has become an important issue in web mining, machine learning and natural language processing. It focus on the rating prediction task. be his or her decision or estimate, the emotional state of the user while writing. Sentiment Analysis is hard Today, Sentiment analysis plays an important role where various machine learning technique is used in determining the sentiment of very huge amounts of text or speech. Various application tasks include such as determining how someone is excited for an upcoming movie, correlates different views for a political party with people’s positive attitude towards vote for that party, or by converting written hotel reviews into 5-star based on scaling across categories like ‘quality of food’, ‘services’, ‘living room’ and ‘facilities’ provided. As there is huge amount of information is shared on social media, forums, blogs, newspaper etc. it is easy to see why there is a need for sentiment analysis as there is much information to process manually which is not possible in today’s time. Text Analysis process Fig 1: An Example of Positive Review and Negative review on websites In Fig.1, we intuitively show an example of positive reviews and negative reviews on website. From Fig.1, there are many positive words in a 5-star review, such as “great”, and “lovely”. But in a 2-star review we find negative words, such as “expensive”, and “poor”. That means a good review reflects a high star-level and a bad review reflects a low-level. When we know the advantages and disadvantages from the two kinds of reviews, we can easily make a decision. Fig 2: Process of Analyzing Text This process involves the following steps [23]: Sentiment Analysis Data Acquisition: In this data acquisition, data are gathered from different relevant sources such as web crawling, twitter tweets, online review, newsfeeds, document scanning etc. Sentiment analysis refers to the use of natural language processing to identify and extract onesided information in source materials or simply it refers to the process of detecting the polarity of the text. It also referred as opinion mining, as it derives the opinion, or the attitude of a user. A common approach of using this is described how people think about a particular topic. Sentiment analysis helps in determining the thoughts of a speaker or a writer with respect to some subject matter or the overall contextual polarity of a document. The attitude may © 2021, IRJET | Impact Factor value: 7.529 Preprocessing: It is used to remove noisy, inconsistent and incomplete data. For doing the classification, Text preprocessing and feature extraction is a preliminary phase. | ISO 9001:2008 Certified Journal | Page 651 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 08 Issue: 04 | Apr 2021 p-ISSN: 2395-0072 www.irjet.net Preprocessing involves 3 steps: identifying is all about to tag the data so it can be create quickly and efficiently. But various organizations can gain from re-transforming their information, which helps in order to cut storage and backup costs, with increasing the speed of data searches. Classification can help an organization to meet authorized and regulatory requirements to retrieve specific information within a specific time period, and this is most important factor behind implementing various data classification technology. Analytical Application: It provides valuable things from text mining so that it can provide information that helps in improving decision and processes. It includes following ways such as sentiment analysis, document imaging, fraud analysis etc. Tokenization or segmentation: It is the process of splitting a string of written language into its words. Text data consists of block of characters referred to as tokens. So the documents are being separated as tokens and have been used for further processing. Removal of stop words: Stop words are the words which are needed to be filtered i.e. may be before or after natural language processing. Stop words are words which contain little informational. Various tools specifically avoid to remove these stop words in order to support phrase search. Several collections of words can be chosen as stop words for any purpose. Some search engines, removes most of the common words which include lexical words such as "want" from a text in order to improve performance. Search engine or natural language processing may contain a variety of stop words. It includes English stop words such as “and”, “the”, “a”, “it”, “you”, “may”, “that”, “I”, “an”, “of” etc. which are considered as ‘functional words’ as they don’t have meaning. II. LITERATURE SURVEY This paper chose primarily three methods for text classification because of their relative popularity and success in prediction of sentiments: Naive Bayes: This works on the assumption of conditional independence and despite this oversimplified assumption, Naive Bayes performs well in many complex real-world problems. Naive Bayes classifier is superior in terms of CPU and memory consumption. Researchers have shown that by removing stop words from the file, you can get the benefit of reduced index size without much affecting the accuracy of a user's. But care should be taken however to take into consideration the user's needs. Mostly, all search engines helps in eliminating the stop words from their indexes. With the help of eliminating stop words from the index, the index size can be reduced to about 33% for a word level index. While assessing the content of natural language processing, meaning of word can be conveyed more clearly by removing the functional word [21]. Stemming: It is the term which used to describe the process to reduce derived words to their origin word stem. Since 1960s, algorithms for stemming have been studied in the field of computer science. Different Stemming methods are commonly referred as stemming algorithms or stemmers. For English, the stemmer example are that, it should identify the string “cats”, ”catty” as based on the root word “cat”, and also “walks”, “walked”, “walking” as based on the root word "walk" [22]. Support Vector Machines: SVM also provides a robust approach to build text classifiers and was picked because of its ability to handle High dimensional input space. When learning text classifiers, many (morethan10000) features can been countered. Since SVMs use over fitting protection, which does not necessarily depend on the number of features, they have the potential to handle these large feature spaces. Maximum Entropy: MaxEnt Naïve Bayes is based on conditional independence assumption, hence to ensure that this paper covers an alternative, it uses Maximum Entropy that does not assume conditional independence. It is based on the Principle of Maximum Entropy and from all the models that fit the training data, selects the one which has the largest entropy. Although it takes more time than Naïve Bayes to train the model, this method has proven to be useful in cases where we do not know anything about the prior distribution (Hening-Thurau etal., 2003) state that customer comments articulated via the Internet are available to a large number of other customer’s, and therefore can be expected to have a significant impact on the success of goods and services. This on consumer buying and communication behavior are tested in a large-scale empirical study. The results illustrate that consumers read online articulations mainly to save decision-making time and make better buying decisions. Structural equation modeling shows that their motives for retrieving online Data Mining: Applying different mining techniques to derive usefulness about stored information. Different mining approaches are classification, clustering, statistical analysis, natural language processing etc. In text analytics, mainly classification technique is used. Classification is a supervised learning method that helps in assigning a class label to an unclassified tuple according to an already classified instance set. Data classifying and © 2021, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 652 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 08 Issue: 04 | Apr 2021 p-ISSN: 2395-0072 www.irjet.net articulations strongly influence their behavior (Duan et al., 2008) showed that both a movie’s box office revenue and WOM valence significantly influence WOM volume. WOM volume in turn leads to higher retrieve other customer’s online articulations from webbased consumer opinion platforms. The relevance of these motives and their impact box office performance. This positive feedback mechanism highlights the importance of WOM in generating and sustaining retail revenue. (Chevalier & Mayzlin, 2006) hypothesized that buyers suspect that many reviewers are authors or other biased parties. They found marginal (negative) impact of 1-star reviews is greater than the (positive) impact of 5-star reviews. The results suggest that new forms of customer communication on the Internet have an important impact on customer behavior. Work on sentiment analysis found using a formal approach is the work by (Simancík and Lee, 2009). The paper presents a method to detect sentiment of newspaper headlines, in fact partially using the same grammar formalism that later will be presented and used in this work, however without the combinatorial logic approach. The paper focus on some specific problems arising with analysing newspaper headlines, e.g. such as headline texts often do not constitute a complete sentence, etc. However the paper also present more general methods, including a method for building a highly covering map from words to polarities based on a small set of positive and negative seed words. This method has been adopted by this thesis, as it solves the assignment of polarity values on the lexical level quite elegantly, and is very loosely coupled to the domain. However, their actual semantic analysis, which unfortunately is described somewhat shallow in the paper, seems to suffer from severe problems with respect to certain phrase structures, e.g. dependent clauses. eWOM is a form of communication, defined as a: “statement made by potential, actual, or former customers about a product or company, which is made available to a multitude of people and institutions via the Internet” (Hennig-Thurau, Gwinner, Walsh , & Gremle, 2004,p. 39). eWOM may be less personal in that it is not face-to-face (or maybe just personal in a different way than in the past), but it is more powerful because it is immediate, has a significant reach, is credible by being in print, and is accessible by others (Hennig-Thurau et al., 2004. same set like, while the item-based CF approach aims to provide a user with the recommendation on an item based on the other items with high correlations (i.e., „neighbours‟ of the item). In all collaborative filtering methods, it is a significant step to find users‟ (or items‟) neighbours, that is, a set of similar users (or items). Currently, almost all CF methods measure users‟ similarity (or items‟ similarity) based on co-rated items of users (or common users of items). Collaborative filtering and content based filtering have been widely used to help users find out the most valuable information. Matrix Factorization based Approaches 1) Basic Matrix Factorization Matrix factorization is one of the most popular approaches for low-dimensional matrix decomposition. Matrix factorization based techniques have proven to be efficient in recommender systems when predicting user preferences from known user-item ratings. Matrix can be inferred by decomposing item reviews that users gave to the items. Matrix factorization methods have been proposed for social recommendation due to their efficiency to dealing with large datasets. several matrix factorization methods have been proposed for collaborative filtering. The matrix approximations all focus on representing the user-item rating matrix with lowdimensional latent vectors. 2) Social Recommendation In real life, people‟s decision is often affected by friends‟ action or recommendation. How to utilize social information has been extensively studied. Yang et al. [6] propose the concept of “Trust Circles” in social network based on probabilistic matrix factorization. Jiang et al. [7] propose another important factor, the individual preference. some websites do not always offer structured information, and all of these methods do not leverage users‟ unstructured information, i.e. reviews, explicit social networks information is not always available and it is difficult to provide a good prediction for each user. For this problem the sentiment factor term is used to improve social recommendation. III. EXISTING SYSTEM Collaborative Filtering Different types of data are generated from different Social media groups that need to be organized and to monitor people’s attitude towards products, gadgets, movie review etc. This database is collected from different social media sites for example Twitter, Face book, Online review, shopping sites etc. Text analytics and Sentiment analysis can help to develop valuable business insights from text based contents that may be in the form of word documents, tweets, comments and news that related to Social media. The foremost reason of Sentiment analysis is Collaborative filtering (CF) is an important and popular technology for recommender systems. The task of CF is to predict user preferences for the unrated items, after which a list of most preferred items can be recommended to users. The methods are classified into user-based CF and item-based CF. The basic idea of user-based CF approach is to find out a set of users who have similar favour patterns to a given user (i.e., „neighbours‟ of the user) and recommend to the user those items that other users in the © 2021, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 653 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 08 Issue: 04 | Apr 2021 p-ISSN: 2395-0072 www.irjet.net so complex is that words often take different meanings and are associated with different emotions depending on the domain in which they are being used. Dataset is analyzed by using the weka tool. The hidden relationship has to be extracted from this type of database using different mining approaches in Weka tool. Dictionary building for detailed sentiment analysis implies making an initial list of adjectives and nouns which are normally used when describing a specific movie review. Phrases and terms are extracted from this relational dataset and their meaning has been added to dictionary for next generation analysis. In tweets, informal and shortcuts has been used for explaining terms or views and this is done with the help of sentiments analysis is not an easy process. To reduce this, data mining approaches has been used for extraction of features from these datasets. large part of reviews are singletons, i.e., there is only one review written by a given reviewer in a certain period of time (this means that in the user account there is only one review at the time of the analysis). For this kind of reviews, specific features must be designed. In fact, as it will be illustrated in the following, many of the features that have been proposed in the literature are based on some statistics over several reviews written by the same reviewer. In the case of singletons, these features loose their relevance in assessing credibility. Therefore, the definition of suitable features that are effective for detecting also singleton fake reviews becomes crucial. 1) Textual Features: as briefly illustrated in Section II, it is practically impossible to distinguish between fake and genuine reviews by only reading their content. The analysis provided by Mukherjee et al. in [19] has shown that the KL-divergence between the languages employed by spammers and non spammers in Yelp is very subtle. However, the good results obtained in [26] by using linguistic features on a domain specific dataset (i.e., a Yelp’s dataset containing only New York japanese restaurants), show that at least on a domain specific level, textual features can be useful. It is possible to use Natural Language Processing techniques to extract simple features from the text, and to use as features some statistics and some sentiment estimations connected to the use of the words.. IV. PROPOSED WORK As briefly introduced in Section II, many and different are the features that have been considered so far in the review site context to identify fake reviews. In some cases, features belonging to different classes have been considered separately by distinct approaches. In other cases, the employed features constitute a subset of the entire set of features that could be taken into account; furthermore, new additional features can be proposed and analyzed to tackle open issues not yet considered, for example the detection of singleton fake reviews. For these reasons, in this section we provide a global overview of the various features that can be employed to detect fake reviews. Both significant features taken from the literature and new features proposed in this article are considered. Since the most effective approaches discussed in the literature are in general supervised and consider reviewand reviewer-centric features, these two classes will be presented in the following sections. The choices behind the selection of the features belonging to the above mentioned classes will be detailed along each section. When the features are taken from the literature, they will be directly referred to the original paper where they have been initially proposed. The absence of the reference will denote those features that have been widely used by almost every proposed technique. Finally, the presence of the label denoted by [new] will indicate a feature proposed for the first time in this article. A. Reviewcentric Features The first class of features that have been considered, is constituted by those related to a review. They can be extracted both from the text constituting the review, i.e., textual features, and from meta-data connected to a review, i.e., metadata features. In every review site, the time information regarding the publication of the review, and the rating (within some numerical interval) about the reviewed business are metadata, are always provided. In addition, in relation to metadata features, those connected to the cardinality of the reviews written by a given user must be carefully studied. In fact, a © 2021, IRJET | Impact Factor value: 7.529 Fig 3: Review Analysis | ISO 9001:2008 Certified Journal | Page 654 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 08 Issue: 04 | Apr 2021 p-ISSN: 2395-0072 www.irjet.net the occurrences of words in whole dataset and forms a dictionary of some most frequently occurring words. V. METHODOLOGY In this proposed system, hadoop open source data mining tool has been used so as to perform sentiment classification on movie review dataset. Here, goal is to classify dataset into positive and negative and form the combined dictionary of Twitter dataset and online review dataset. Main steps are: The online review dataset consists of around 800 user’s review archived on the IMDB (Internet Movie Database) portal. And for, Twitter dataset around 1000 review were collected and each review were formatted according to .arff file where review text and class label are only two attributes. Here, we analyse the dataset based on accuracy given by naïve bayes multinomial. Online review dataset accuracy around 94.968% and for twitter its around 82.695%. Results show that we get better accuracy for online review as compared to twitter tweets as online review are more clear and in detail compare to twitter tweets. Generating Dataset Two dataset were collected firstly, from Twitter tweets and secondly, from Online review Dataset. The online review dataset consists of around 800 user’s review archived on the IMDB (Internet Movie Database) portal. And for, Twitter dataset around 1000 review were collected and each review were formatted according to .arff file where review text and class label are only two attributes. Class label represent the overall user opinion. Here, we set simple rules for scaling the user review. For dataset, a user rating greater than 6 is considered as positive, between 4 to 6 considered as neutral and less than 4 considered as negative. Preprocessing For doing the classification, Text preprocessing and feature extraction is a preliminary phase. Preprocessing involves 3 steps: I. Word parsing and tokenization: In this phase, each user review splits into words of any natural processing language. As movie review contains block of character which are referred to as token. II. Removal of stop words: Stop words are the words that contain little information so needed to be removed. As by removing them, performance increases. Here, we made a list of around 320 words and created a text file for it. So, at the time of preprocessing we have concluded this stop word so all the words are removed from our dataset i.e. filtered. Fig 4: Performance analysis based on accuracy Combined dictionary of words of twitter tweets and online review are formed based on probability of each word as we get by classification algorithm i.e. naïve bayes multinomial. III. Stemming: It is defined as a process to reduce the derived words to their original word stem. For example, “talked”, “talking”, “talks” as based on the root word “talk”. We have used Snowball stemmer to reduce the derived word to their origin. VI. CONCLUSION Determining and categorizing reviews to be false or factual is an important and challenging issue. In this paper, we have used language features such as presence of unigram, frequency of unigram, presence of bigram, frequency of bigram and length of reviews to build the model and detect false reviews. After applying the above model we have come to the conclusion that, obtaining false reviews requires both linguistic and behavioral features This paper focuses on obtaining confidential reviews using supervised reading on language features only. The same model can also be started by combining behavioral and linguistic features using supervised, uncontrolled, or Classification Classification is a supervised learning method that helps in assigning a class label to an unclassified tuple according to an already classified instance set. Here, naïve bayes multinomial classifier has been used. Quality measure will be considered on the basis of percentage of correctly classified instances. For the validation phase, we use 10fold cross validation method. Naïve bayes multinomial helps in generating dictionary and frequent set. It counts © 2021, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 655 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 08 Issue: 04 | Apr 2021 p-ISSN: 2395-0072 www.irjet.net [11] T. Nakagawa, K. Inui, and S. Kurohashi, “Dependency tree-based sentiment classification using CRFs with Hidden Variables,” NAACL, 2010, pp.786-794. supervised learning methods. REFERENCES [1] B. Wang, Y. Min, Y. Huang, X. Li, F. Wu, “ Review rating prediction based on the content and weighting strong social relation of reviewers,” in Proceedings of the 2013 international workshop of Mining unstructured big data using natural language processing, ACM. 2013, pp. 23-30. [12] Xiaojiang Lei, Xueming Qian, Member, IEEE, and Guoshuai Zhao, “Rating Prediction based on Social Sentiment from Textual Reviews,” IEEE Transactions On Multimedia, MANUSCRIPT ID: MM-006446 [13] Mrs. R.Nithya, Dr. D.Maheshwari. 2014 Sentiment Analysis on Unstructured Review, International Conference on Intelligent Computing Application, IEEE. [2] D. Tang, Q. Bing, T. Liu, “Learning semantic representations of users and products for document level sentiment classification,” in Proc. 53th Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing, Beijing, China, July 26-31, 2015, pp. 1014– 1023. [14] Ms. K. Mouthami, Ms. K.Nirmala Devi, Dr. V.Murali Bhaskaran. 2010 Sentiment Analysis and Classification based on Textual Reviews, Dept of CSE, Tamil Nadu, IEEE. [15] Nargiza Bekmamedova, Graeme Shanks 2013 Social Media Analytics and Business Value: A Theoretical Framework and Case Study, Department of Computing and Information Systems, University of Melbourne. [3] Y. Zhang, G. Lai, M. Zhang, Y. Zhang, Y. Liu, S. Ma, “Explicit factor models for explainable recommendation based on phrase-level sentiment analysis,” in proceedings of the 37th international ACM SIGIR conference on Research & development in information retrieval, 2014. [16] Simona Vinerean, Iuliana Cetina 2013 The Effects of Social Media Marketing on Online Consumer Behavior, International Journal of Business and Management; Vol. 8, No. 14. [4] W. Zhang, G. Ding, L. Chen, C. Li , and C. Zhang, “ Generating virtual ratings from Chinese reviews to augment online recommendations,” ACM TIST, vol.4, no.1. 2013, pp. 1-17. [17] SitaramAsur, Bernardo A.Huberma 2012 Predicting the Future with Social Media, Social Computing Lab, HP Labs, Palo Alto, California. [5] X. Lei, and X. Qian, “Rating prediction via exploring service reputation,” 2015 IEEE 17th International Workshop on Multimedia [18] V.K. Singh, R.Piryani, A. Uddin, P.Waila. 2013 Sentiment Analysis of Movie Reviews, Department of Computer Science, New Delhi, India, Published in IEEE. [6] X. Yang, H. Steck, and Y. Liu, “Circle-based recommendation in online social networks, ” in Proc. 18th ACM SIGKDD Int. Conf. KDD, New York, NY, USA, Aug. 2012, pp. 1267–1275. [19] V. S. Jagtap, Karishma Pawar. 2013 Analysis of different approaches to Sentence-Level Sentiment Classification, published in IJSET. [7] M. Jiang, P. Cui, R. Liu, Q. Yang, F. Wang, W. Zhu, and S. Yang, “Social contextual recommendation,” in proc. 21st ACM Int. CIKM, 2012, pp. 45-54. [20] Ya-Ting Chang, Shih-Wei Sun 2013 A Real time Interactive Visualization System for Knowledge Transfer from Social Media in a Big Data, Center for Art and Technology, Taipei National University of the Arts, Taipei, Taiwan, IEEE. [8] Z. Fu, X. Sun, Q. Liu, et al., “Achieving Efficient Cloud Search Services: Multi-Keyword Ranked Search over Encrypted Cloud Data Supporting Parallel Computing,” IEICE Transactions on Communications, 2015, 98(1):190200. [21] Stop http://en.wikipedia.org/wiki/Stop_words words: [22] Stemming: http://en.wikipedia.org/wiki/Stemming [9] Y. Ren, J. Shen, J. Wang, J. Han, and S. Lee, “Mutual Verifiable Provable Data Auditing in Public Cloud Storage,” Journal of Internet Technology, vol. 16, no. 2, 2015, pp. 317-323. [23] Decideo Amérique Du Nord www.decideo.fr/bruley/docs/2sentiment_a_v0.ppt Inc [10] W. Luo, F. Zhuang, X. Cheng, Q. H, Z. Shi, “Ratable aspects over sentiments: predicting ratings for unrated reviews,” IEEE International Conference on Data Mining (ICDM), 2014, pp. 380-389. © 2021, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 656