Abstract: In the past decades, people have gained a wide range of options as the availability of information expands. To help them make decisions, recommendation systems play an important role in all kinds of aspects, e.g. news, books, movies and so on. One such aspect is Restaurant where recommendations can be provided using user attributes and past activity. A noticeable similarity is found in people belonging to same categories based on attributes like age, native place, gender, work-type, etc. Using Collaborative approach these attributes of individuals can be analysed. Also the reviews and ratings given by customers to a restaurant play an important role in selection of an ideal restaurant. In this paper, we follow an approach based on the Simple Bayesian Classifier and apply it to user-based variant of the collaborative filtering, which makes predictions based on the user similarities. The recommended results are further refined by the review/rating analysis of individual restaurants using Text Mining. The review/rating analysis of predicted restaurants help to assess the current overall user experience of those restaurants which include the quality of food served, service, cost, ambience, etc. Our approach comprises counting positive and negative term scores to determine sentiment orientation, using Sentiment Analysis (SentiWordNet library). Finally more relevant results with positive reviews can be obtained which are passed as output recommendations to customers. In future we can also add content based filtering to recommend restaurant on the basis of the characteristics like dinning arrangement, facilities, working hours, etc. of restaurants that the particular user have already visited. By making hybrid of both content and collaborative we can increase the quality of recommendation result.
Keywords: Recommender System, Collaborative filtering, Naive Bayes, SentiWordNet.