ISSN:2229-6093

Partha Sarathi Chakraborty et al, International Journal of Computer Technology & Applications,Vol 7(3),477-483

Better Experience of Favourite Food using Sentimental Analysis Partha Sarathi Chakraborty, Arti Tiwari Assistant Professor, Department of CSE, SRM University, India [email protected], [email protected]

ABSTRACT This paper analyses the customer reviews on restaurant domain using sentiment analysis and text mining techniques. It allows user to search a particular food item within their own budget and select a place to dine-in with the review generated by analyzing the sentiments. The most integral part of our work is to assign Sentiment scores to the aspects with respect to the words used. We have devised Naïve Bayes algorithm to perform this function. The dataset we have at our disposal is a set of review documents obtained from an authenticated repository. We perform an aspect level sentiment extraction thereby, attempting to mine and understand the user's feedback data. The aspects that we have taken into account are food, cost, ambience and service. The algorithm forms the rule base for the classifier to predict the polarity of the reviews. To start with, we perform a cleanup on the review data. Polarity prediction is performed using Naïve Bayes and the results are evaluated. Finally, the experimental analysis shows that, Naïve Bayes performs better with increasing number of instances. The proposed methodology will equip the restaurant owner to identify the areas that require improvement.

General Terms Favourite Food, Sentimental Analysis.

Keywords Naïve Bayes, Polarity, Aspect of Restaurant Domain, Text Mining.

1. INTRODUCTION Information is generated and managed by one person and is consumed by many other persons, with most of this usergenerated content being textual information. Typing of text query on mobile device and digesting a large number of retrieved user-generated data are both inconvenient for users. As there is a lot of raw data from real time messages posted by people about their opinions on a variety of topics in daily life, it is a worthwhile research endeavour to collect and analyse this data, which may be useful for users or managers to make informed decisions.

1.1 Overview of Sentimental Analysis Sentiment analysis and opinion mining is the field of study that analyses people's opinions, sentiments, evaluations, attitudes, and emotions from written language. It is one of the most active research areas in natural language processing and is also widely studied in data mining, Web mining, and text mining. In fact, this research has spread outside of computer science to the management sciences and social sciences due to its importance to business and society as a whole. The growing importance of sentiment analysis coincides with the growth of social media such as reviews, forum discussions,

IJCTA | May-June 2016 Available [email protected]

blogs, micro-blogs, Twitter, and social networks. Sentiment analysis provides easy-to-use mechanisms to identify the positive or negative sentiment within any document or webpage. The sentiment analysis works on documents large and small, including news articles, blog posts, product reviews, comments and Tweets. Sentiment analysis (also known as opinion mining) refers to the use of natural language processing, text analysis and computational linguistics to identify and extract subjective information in source materials. Generally speaking, sentiment analysis aims to determine the attitude of a speaker or a writer with respect to some topic or the overall contextual polarity of a document. The attitude may be his or her judgment or evaluation, affective state (i.e., the emotional state of the author when writing), or the intended emotional communication (i.e., the emotional effect the author wishes to have on the reader). A basic task in sentiment analysis is classifying the polarity of a given text at the document, sentence, or feature/aspect level - whether the expressed opinion in a document, a sentence or an entity feature/aspect is positive, negative, or neutral. Microblogging websites have evolved to become a source of a diverse variety information, with millions of messages appearing daily on popular web-sites. Users can post real time messages about their life, share opinions on variety of topics and discuss current issues on these microblogging websites. Product reviewing has been rapidly growing in recent years because more and more products are selling on the Web. The large number of reviews allows customers to make informed decisions on product purchases. However, it is difficult for product manufacturers or businesses to keep track of customer opinions and sentiments on their products and services. In order to enhance the customer shopping experiences a system is needed to help people analyze the sentiment content of product reviews. The power of social media as a marketing tool has been recognized, and is being actively taken advantage of by people, governments, major corporations, and schools. Twitter, Facebook are perhaps the most popular microblogging website where users create status messages called “tweets” and “status”, which are updates and musings from users that must be written in 140 characters or less. Tweets containing opinions are important because whenever people need to make a decision, they want to hear others’ opinions. The same is true for organizations. However, many real-life applications require very detailed analyses in order to gather information from, for example, a product review, whose data could help users or managers make important product-related decisions. This approach is also being actively employed by governments or companies to collect and analyse feedback on their policies or products.

477

ISSN:2229-6093

Partha Sarathi Chakraborty et al, International Journal of Computer Technology & Applications,Vol 7(3),477-483

1.2 Sentimental Analysis on Food In recent years, food has become the rising star of public debate. Documentaries such as Food Inc, Fast Food Nation, King Corn and Supersize Me, have brought food issues into the realm of popular culture and into sharp focus for many urbanites. Food consumption – a naturally social phenomenon – and its reflection in the emotional social web of Twitter, Facebook and Restaurant Blogs becomes a lens to reveal patterns in society. The ever-increasing popularity of websites that feature user-generated opinions (e.g., Zomato.com, FoodSpotting.com) has led to an abundance of customer reviews that are often too numerous for a user to read. Arguably, government and citizen interest, consumer focus and the food industry itself are at an enormous tipping point. Sentimental Analysis is a data visualization tool that explores the opportunities presented by the data-sharing world of today’s cities using global English-language reviews about food. It aims to gain a better understanding of global food consumption patterns and its impact on the daily emotional well-being of people. Consequently, there is a growing need for systems that are able to automatically extract, evaluate and present opinions in ways that are both helpful and easy for a user to interpret. The relationship between language and sentiment is an active area of investigation. Much of this research has focused on customer-written reviews of goods and services, and perspectives have been gained on how sentiment is expressed in this type of informal text. In addition to sentiment, however, other variables are reflected in a reviewer’s choice of words, such as the price of the item under consideration. Sentiment Analysis can used to extract sentiment from the restaurant blogs about food. Sentiment analysis is an application of natural language processing, computational linguistics, and text analytics to identify and extract subjective information from different source materials. One of the most basic tasks of the sentiment analysis tool is to determine the overall tonality and classify the polarity of a given text to indicate whether a sentence or a feature of a document is positive, negative, or neutral. Using a method similar to Sentiment Analysis Classifier for Twitter, the classifier was trained on one million tweets in the food domain (2014). In the research by Go et al accuracies of 83% are achieved. Go et al describes that it is not really possible to train a classifier to perform better than 85% accuracy for Sentiment Analysis tasks, as tests have shown that human annotators cannot agree on the right classification in about 15% of the cases (2014).Naive Bayesian classifier reached similar accuracy levels within the food domain as the linguistic form of the source materials was the same (tweets about food). By using the Bayesian Sentiment classifier the exact probability that the review is positive (or negative) is returned to the system. We use the positive Bayesian chance as a "happiness percentage". General architecture called Opinion Miner combines various supervised learning methods to improve the final accuracy of sentiment classification. Because the labelling of data is very time- consuming, many researchers use data which contains emoticons to identify sentiment, and use these data as training data. However, since emoticons are not always consistent with the sentiment, there will be many mistakes in the training data. Thus, this kind of opinion miner helps to investigate the text from the social media to classify them as per the need of an organisation. Therefore, manually labelled data is used as the training data to build a model shown in fig 1.

IJCTA | May-June 2016 Available [email protected]

Fig. 1: Opinion Miner Architecture In Pre-processing, first we will introduce various properties of messages that users post on Blogs. Second we eliminate reviews that are not in English, have just a URL, remove tags. In extraction of Reviews containing Opinions, the real world reviews containing opinion is more valuable. So in this part we want to filter out the reviews without opinion. To do this, we use Naive Bayes (NB) classifier on the training that data we labelled manually to classify reviews composed of opinions or non-opinions. Naive Bayes is a simple model which works well on text categorization. In Short Text Classification, the intrinsic idea of this part is that we observed that a word may have different meanings in different domains. For example, “@htc I wouldn't know, no beats headphones came with my beats device :( sad” and “@teresa_fnts YES! they were. So sad that was the last twilight movie :( I loved wreck it Ralph. I took my nephew to see it :) he loved it!”. We can see that “sad” means a negative word in the former example and the “sad” in the latter means a positive word. So the unigram Naïve Bayes classifier method is used together with the pre-labelled training data to build the multi-classifier. We use distinct categories of training data. By Training Multiple Classifiers in Distinct Categories we now reach the step of predicting the orientation of an opinion sentence, i.e., positive or negative. As we mentioned above, some words in different areas or categories can have different meanings. In order to improve the final identification accuracy, we need to first classify the short texts according to their domains, so that the classifier can automatically classify with greater performance the reviews as being either positive or negative. By using these training data and Naive Bayes method, we can build many binary classifiers of different categories to complete the system. However since labelling data is very time-consuming, the size of this training data is small.

2. Literature Survey Gediminas Adomavicius et. al. [1] proposed New Recommendation Techniques for Multicriteria Rating Systems that explain personalization technologies and recommender systems help online consumers avoid information overload by making suggestions regarding which information is most relevant to them. Most online shopping sites and many other applications now use recommender systems. The vast majority of current recommender systems use a single criterion, such as a single numerical rating, to represent an item’s utility to a user in a 2D Users-Items space. Single-criterion rating systems have proved successful in several applications, but many industries have begun

478

ISSN:2229-6093

Partha Sarathi Chakraborty et al, International Journal of Computer Technology & Applications,Vol 7(3),477-483 employing multicriteria systems. For example, restaurant guides, such as Zagat’s Guide, provide three criteria for restaurant ratings (food, décor,and service). This move indicates that multicriteria data provides value to online content providers and consumers as a component in personalization applications. Taking full advantage of multicriteria ratings in personalization applications requires new recommendation techniques. In this article, we propose several new techniques for extending recommendation technologies to incorporate and leverage multicriteria rating information. To evaluate experimental results of the proposed approaches, we collected a set of user-submitted movie ratings from the Yahoo! Movies Web site (http://movies.yahoo.com) for several hundred randomly chosen movies from the last decade. When users submit movie ratings to Yahoo! Movies, in addition to the overall rating, they are asked to provide four criteria information for each movie: story, acting, direction, and visuals. All ratings have 13 possible values and are based on a standard grading scale from A+ to F; for our analysis, we changed the grades to numerical values from 13 to 1. In the data pre-processing stage, we invoked two constraints on the data set to ensure that it was not extremely sparse and had sufficient data for rating prediction: that each user rated at least 10 movies and that each movie had at least 10 user ratings. The resulting data set included 155 users, 50 movies, and 2,216 known ratings in total. Limitation of this approach in terms of the performance of a specific technique is highly domain-dependent; it depends significantly on the underlying data’s characteristics. So, although we expect the proposed techniques to do well in a variety of application domains, we do not expect them to outperform traditional single-criterion techniques in all domains where multicriteria information exists. For example, these techniques cannot be expected to perform well in domains where multicriteria ratings do not carry meaningful information or where no inherent relationship exists between the overall rating and the multicriteria ratings for the users or items. Po-Wei Liang et. al. [2] discuss Opinion Mining on Social Media Data. through introduce Microblogging. Microblogging websites have evolved to become a source of a diverse variety information, with millions of messages appearing daily on popular websites. Users can post real time messages about their life, share opinions on variety of topics and discuss current issues on these microblogging websites. Product reviewing has been rapidly growing in recent years because more and more products are selling on the Web. However, it is difficult for product manufacturers or businesses to keep track of customer opinions and sentiments on their products and services. In order to enhance the customer shopping experiences a system is needed to help people analyse the sentiment content of product reviews. Most of the traditional research has focused on classifying long texts, such as reviews. However since Microblogging messages that are short and colloquial, traditional algorithms do not perform as well as they do for long texts. For this reason, there has been in recent years a lot of research in the area of sentiment classification that has targeted microblogging data In this paper, to overcome these challenges, they aim to design a system which automatically combines supervised learning that is capable of extracting, learning and classifying tweets, with opinion expressions. The basic idea is to use

IJCTA | May-June 2016 Available [email protected]

domain-specific training data to build a generic classification model from social media data to help improve the performance. The experimental results demonstrate the effectiveness of the proposed system is work well. They propose new system architecture and combine various supervised learning methods to improve the final accuracy of sentiment classification and to automatically analyze the sentiments of these messages. They determine the data’s category first, because we assume that different domains are associated with different customary terms and expressions, which will affect the accuracy of sentiment analysis. They combine this system with manually annotated data from Twitter, one of the most popular microblogging platforms, for the task of sentiment analysis. In this system, machines can learn how to automatically extract the set of messages which contain opinions, filter out non opinion messages and determine their sentiment directions (i.e. positive, negative). Proposed system also explored linguistic relationships between food prices and customer sentiment through quantitative analysis of a large datasets of menus and reviews. We have also proposed visualization techniques to better understand what our models have learned and to see how they can be applied to new data. More broadly, this paper is an example of using extrinsic variables to drive model-building for linguistic data, and future work might explore richer extrinsic variables toward a goal of task-driven notions of semantics. Victor Chahuneau et. al. [4] proposed a method that builds on a thread of NLP research that seeks linguistic understanding by predicting real-world quantities from text data. Prediction tasks are used on language use on restaurant menus and in user restaurant reviews for quantitative evaluation and objective model comparison, while analysis of learned models gives insight about the social process behind the data. It uses data from a large corpus of restaurant menus and reviews crawled from the web and formulate several prediction tasks. In addition to predicting menu prices, we also consider predicting sentiment along with price. They consider several prediction tasks like predicting individual menu item prices, predicting the price range for each restaurant and finally jointly predicting median price and sentiment for each restaurant. Hence this paper contributes an exploratory data analysis of language used to describe food (by its purveyors and by its consumers). While their primary goal is to understand the language used in our corpus, our findings bear relevance to economics and hospitality research as well. This paper is a step on the way to the eventual goal of using linguistic analysis to understand social phenomena like sales and consumption. Yohan Jo et. al. [5] first propose Sentence-LDA (SLDA), a probabilistic generative model that assumes all words in a single sentence are generated from one aspect. They then extend SLDA to Aspect and Sentiment fication Uni Model (ASUM), which incorporates aspect and sentiment together to model sentiments toward di fferent aspects. ASUM discovers pairs of {aspect, sentiment} which we call senti-aspects. They applied SLDA and ASUM to reviews of electronic devices and restaurants. The results show that the aspects discovered by SLDA match evaluative details of the reviews, and the senti-aspects found by ASUM capture important aspects that are closely coupled with a sentiment. The results of sentiment

479

ISSN:2229-6093

Partha Sarathi Chakraborty et al, International Journal of Computer Technology & Applications,Vol 7(3),477-483 classification show that ASUM out- performs other generative models and comes close to super- vised classi fication methods. One important advantage of ASUM is that it does not require any sentiment labels of the reviews, which are often expensive to obtain. They applied their models to various tasks, and the experiments show that our models perform well in Aspect discovery, Senti-aspect discovery, Aspect-specific sentiment words and Sentiment classification. In the quantitative evaluation of sentiment fication, classi ASUM outperformed other generative models and came close to supervised classification methods.

3. PROBLEM DESCRIPTION 3.1 Existing Problem People make decisions every day. “Which movie should I see?”, “Which city should I visit?”, “What should I eat?”... There are too many choices and a little time to explore them all. The problem most users face nowadays is the lack of time; most people are unable to read the reviews and just rely on the business’ ratings. Ratings are useful to convey the overall experience. In case of a restaurant, no existing site spot dishes based on the price of an individual dish in collaboration with the food quality.

Fig. 2: Post Responses

knowledge relating to food issues. The survey clearly shows that around 85% of the users are concerned about their cash, and prefers it as their major factor while dine-in to a place. In collaboration with the price, the survey brought us a clear picture that users are equally concerned about their health and consider food quality as another major factor. Further, while analysing different existing sites, we collected information that user face a problem to spot the particular dish in their locality with available budget. The pie-chart depicts that user majorly posts their responses on Facebook, Restaurant Blogs and refer these before visiting a new place to dine-in. Thus, this study helps us to conclude that the reviews are to be extracted from Facebook, Restaurant Blogs to examine the sentiments of the users. The graph models to conclude that the common problem that arouse among people was that they cannot spot dishes with the “price” of a particular food item in collaboration with the food quality.

3.3 Proposed Solution The proposed solution will derived people sentiments (views) from different online social networking, microblogging services, online restaurant blogs(like Zomato.com). The software will perform an automatic analysis of different posts, messages, etc. to detect messages of persons, companies or brands (including ambiguous or misspelled names), identify topics and analyze opinions (sentiment). It (Sentiment analysis) has many other names also, including • Opinion Extraction • Sentiment Mining There are three main classification levels in SA: • document-level SA • sentence-level SA • aspect-level SA Document-level SA aims to classify an opinion document as expressing a positive or negative opinion or sentiment. It considers the whole document a basic information unit under a particular domain, here, reviews of particular dishes from restaurant blogs. Document-level SA has important implications for text processing applications such as information extraction and information retrieval. Our software, FoodSpot, includes the search of a food item within the given budget of the user. It extracts the document from Zomato.com for each particular food item and Sentiment Mining is done on the reviews using Naïve Bayesian Algorithm to detect the polarity of the particular food item in terms of quality of the food. The software then produces desired result. Marcel Blattner et. al. [3] build on a well-documented influence of social interactions with peers on the decision to vote, favour, or even purchase an item. A threshold mechanism is used to govern the decision making process that determines whether a user is or is not interested in an item.

Fig. 3: Top Priorities

3.2 Analysis of Problem Domain From burgers and fries to burritos and pizza, the options are nearly limitless. With all the choices out there, it can be difficult to determine how to spend your hard-earned cash. The Food survey done among people was found useful to collect information about reported behaviours, attitudes and

IJCTA | May-June 2016 Available [email protected]

Our model is formulated within an opinion formation framework where social ties play a major role. The main ingredients of our model:fluence In -Network (IN), Intrinsic-ItemAnticipation (IIA), and Influence -Dynamics (ID). It firstly describe how individuals’ Intrinsic - ItemAnticipations may change due to social interactions taking place on a particular fluence In -Network. Secondly, we introduce dynamical processes governing the opinion propagation. This model is inspired by opinion formation taking place on a complex network with a predefined topology. The model is

480

ISSN:2229-6093

Partha Sarathi Chakraborty et al, International Journal of Computer Technology & Applications,Vol 7(3),477-483 able to generate data observed in real world recommender systems. Despite its simplicity, the model isfl exible enough to generate a wide range fferent of patterns. di We mathematically analyse the model using field a mean approach to the full Master Equation. The approach provides an understanding of the data in recommender systems as a product of social processes. The model can serve as a data generator which is valuable for testing and evaluation purposes for recommender systems.

4. METHODOLOGY 4.1 Existing Methods Table 1 shows the net outcome of our comparative study based on various features. It shows the existing methods used for Opinion Mining. Table 1. Differential Analysis Features

Naive Bayes

Max Entropy

Boosted Trees

Random Forest

Based on

Bayes Theorem

Feature Based Classifie r

Decision Tree Learning

Decision Tree Aggregati on

Simplicity

Very Simple

Hard

Moderat e

Simple

Performan ce

Better

Good

Good

Excellent

Accuracy

Good

High

Poor

Excellent

Memory Requireme nt

Low

High

Low

High

Other Applicatio ns

Spam Detectio n, Docume nt Classific ation, Sexually Explicit Content Detectio n

Diagnosi s Tests in Patholog y Labs

Classifyi ng Cardiova scular Outcome s

Biomedica l Applicatio ns, Sexually Explicit Content Detection

Result Accuracy over a period of Time

Variable

Consiste nt

Incremen tal

Increment al

Time Required for Training Classifier

Less

4.2 Proposed Algorithm Naive Bayes has been studied extensively since the 1950s. It was introduced under a different name into the text retrieval community in the early 1960s, and remains a popular (baseline) method for text categorization, the problem of judging documents as belonging to one category or the other (such as spam or legitimate, sports or politics, etc.) with word frequencies as the features. Bayesian classifiers are based around the Bayes rule, a way of looking at conditional probabilities that allows you to flip the condition around in a convenient way. Naive Bayes classifiers are highly scalable, requiring a number of parameters linear in the number of variables (features/predictors) in a learning problem. A conditional probably is a probably that event X will occur, given the evidence Y. That is normally written P(X | Y). The Bayes rule allows us to determine this probability when all we have is the probability of the opposite result and of the two components individually: P(X | Y) = P(X) P(Y | X) / P(Y). This restatement can be very helpful when we're trying to estimate the probability of something based on examples of it occurring. Formula looks like this. P(Sentiment|Sentence)

= P(Sentiment)P(Sentence|Sentiment) ÷ P(Sentiment)

In this case, we're trying to estimate the probability that a document is positive or negative, given its contents. We can restate that so that is in terms of the probability of that document occurring if it has been predetermined to be positive or negative. This is convenient, because we have examples of positive and negative opinions from our data set above. The thing that makes this a "Naïve" Bayesian process is that we make a big assumption about how we can calculate at the probability of the document occurring: that it is equal to the product of the probabilities of each word within it occurring. This implies that there is no link between one word and another word. This independence assumption is clearly not true: there are lots of words which occur together more frequently that either does individually, or with other words, but this convenient fiction massively simplifies things for us, and makes it straightforward to build a classifier. We can estimate the probability of a word occurring given a positive or negative sentiment by looking through a series of examples of positive and negative sentiments and counting how often it occurs in each class. This is what makes this supervised learning - the requirement for pre-classified examples to train on. We can drop the dividing P(line), as it's the same for both classes, and we just want to rank them rather than calculate a precise probability. We can use the independence assumption to let us treat P (sentence | sentiment) as the product of P (token | sentiment) across all the tokens in the sentence. So, we estimate P (token | sentiment) as Count (this token in class) + 1 / count (all tokens in class) + count (all tokens)

Moderat e

IJCTA | May-June 2016 Available [email protected]

High

Recurrent Learning with every novel dataset

The extra 1 and count of all tokens is called 'add one' or Laplace smoothing, and stops a 0 finding its way into the multiplications. If we didn't have it any sentence with an unseen token in it would score zero. The classify function starts by calculating the prior probability (the chance of it being one or the other before any tokens are looked at) based on the number of positive and negative examples - in this example that'll always be 0.5 as we have the same amount of data for each. We then tokenize the incoming document, and for each class multiply together the likelihood of each word being seen in that class. We sort the final result, and return the highest scoring class. 481

ISSN:2229-6093

Partha Sarathi Chakraborty et al, International Journal of Computer Technology & Applications,Vol 7(3),477-483 The Bayes Naïve Classifier selects the most likely classification Vnb given the attribute values a1, a2…….an. This results in: n= the number of training examples for which v = vj ne = number of examples for which v = vj and a=ai p = a priori estimate for P( ai | vj ) m = the equivalent sample size. A naive Bayes classifier assumes that the presence of a particular feature of a class is unrelated to the presence of any other feature. For example, a person may be considered to be a male if he is tall, has short hair and a strong build. Even if these features depend on each other or upon the existence of the other features, a naïve Bayes classifier considers all of these properties to independently contribute to the probability that the person is a male.

Fig. 6: List of Restaurants This page displays the list of available food items in the given budget by the user along with the rating.

Naïve Bayes classifier assumes that the classes for classification are independent. There are various proofs that show that even though the probability estimates of Naïve Bayes classification are low it delivers quite good results in real life examples. Naïve Bayes just over estimates the class that certain object belongs too. Assuming that we are using it only for making decisions (which is true in case of sentiment analysis problem) the decision making is correct and the model is useful.

5. Implementation Figure 4 displays the food items that user can select. The search engine contain two parameters: Food name and Price. This Page also contains Administration login to redirect to admin page where dataset is created.

Fig. 4. Home Page of FoodSpot

Fig. 7. Training Dataset On clicking “Submit” button, this page trains the data available on restaurant blogs to create the dataset.

Fig. 8: Dataset created after Training

6. CONCLUSION

Fig. 5. Search Engine This page displays a popup price search engine, on “submit” redirects to another page.

IJCTA | May-June 2016 Available [email protected]

The Naïve Bayes classifier clearly has an upper hand with high accuracy and performance, simplicity in understanding, and improvement in results over a period of time. This makes the classifier best fit for situations like sentiment analysis. Naïve Bayes is also very well supported in terms of implementation. If processing power and memory is an issue then the Naïve Bayes classifier should be selected due to its low memory & processing power requirements. Selection of a classification model should be done wisely in sentiment analysis systems because this decision will influence the precision of your system and your end product.

482

ISSN:2229-6093

Partha Sarathi Chakraborty et al, International Journal of Computer Technology & Applications,Vol 7(3),477-483 FoodSpot is the software that provides an effective result to the user. It allows user to select the best food location within their budget. The classifier used extract the reviews to show the rating on scale of five. FoodSpot uses large amount of information from platforms like Zomato.com to make them viable for use as data sources, in applications based on opinion mining and sentiment analysis. FoodSpot helps the user to avoid reading thousands of reviews regarding each particular food item. It thus provide a good piece of information to the end users.

7. FUTURE WORK Media information plays a great role in expressing people’s feelings, or opinions about a certain topic or product. Using social network sites and micro-blogging sites as a source of data still needs deeper analysis. There are some benchmark data sets especially in reviews which are used for algorithms evaluation. In many applications, it is important to consider the context of the text and the user preferences. That is why we need to make more research on context-based SA.

8. REFERENCES

[7] I-Chun Liu, I-Chun Chen, Ming-Syan Chen, "Le Festin: Shop sign recognition assisted food recommendation system", International Symposium on Wearable Computers (ISWC), IEEE Sponsored Conference, Seoul, 10-13 Oct. 2010, pp. 1 - 8 [8] Andranik Tumasjan, Timm O. Sprenger, Philipp G. Sandner, Isabell M. Welpe "Predicting Elections with Twitter: What 140 Characters Reveal about Political Sentiment", Proceedings of the Fourth International AAAI Conference on Weblogs and Social Media 2010, 23 - 26 May 2010, George Washington University, pp. 178 - 185 [9] Yoosin Kim, Do Young Kwon and Seung Ryul Jeong “Visualizing the Results of Opinion Mining from Social Media Contents: Case Study of a Noodle Company”, Graduate School of Busi2ness IT, Kookmin University, J Intell Inform Syst 2014 December: 20(4), pp. 89 - 105, http://dx.doi.org/10.13088/jiis.2014.20.4.105 [10] Anthony Aue and Michael Gamon, “Customizing Sentiment Classifiers to New Domains: a Case Study”, Microsoft Research 1, Microsoft Way Redmond, WA. USA

[1] Adomavicius, G., Young Ok Kwon, "New Recommendation Techniques for Multicriteria Rating Systems", Journal of Intelligent Systems, IEEE, Volume:22 Issue 3, June 2007, pp. 48 – 55 [2] Po-Wei Liang, Bi-Ru Dai, "Opinion Mining on Social Media Data", IEEE 14th International Conference on Mobile Data Management (MDM), 2013 (Volume:2 ), Milan, 3-6 June 2013, pp. 91 – 96 [3] Marcel Blattner, Matus Medo, "Recommendation systems in the scope of opinion formation: a model", Proceedings of the 2nd Workshop on Human Decision Making in Recommender Systems in conjunction with the 6th ACM Conference on Recommender Systems (RecSys 2012), Dublin, Ireland, September 9, 2012, pp. 32 - 39, http://ceur-ws.org/Vol-893/paper6.pdf [4] Victor Chahuneau, Kevin Gimpel, Bryan R. Routledge, "Word salad: relating food prices and descriptions", Lily Scherlis, Noah A. Smith, Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning (Proceeding EMNLP-CoNLL '12), 2012, pp. 1357-1367, http://dl.acm.org/citation.cfm?id=2391102 [5] Yohan Jo, Alice H. Oh,"Aspect and sentiment unification model for online review analysis", Proceedings of the fourth ACM international conference on Web search and data mining 2011, (Proceeding WSDM '11), pp. 815-824, http://dl.acm.org/citation.cfm?id=1935932 [6] Deuk Hee Park, Hyea Kyeong Kim, Il Young Choi, Jae Kyeong Kim, "A literature review and classification of recommender systems Research", Journal of Expert Systems with Applications, Elsevier, Volume 39, Issue 11, 1 September 2012, pp. 10059–10072

IJCTA | May-June 2016 Available [email protected]

483