International Journal of Computer Applications (0975 – 8887) Volume 87 – No.14, February 2014 33 Social Popularity based SVD++ Recommender System Rajeev Kumar Student, M. Tech IET, Alwar, Rajasthan, India B. K. Verma, Ph.D Associate Prof., CS Deptt. IET, Alwar, Rajasthan, India Shyam Sunder Rastogi Student, M. Tech IET, Alwar, Rajasthan, India ABSTRACT Recommender systems have shown a lot of awareness in the past decade. Due to their great business value, recommender systems have also been successfully deployed in business, such as product recommendation at flipkart, HomeShop18, and music recommendation at Last.fm, Pandora, and movie recommendation at Flixstreet, MovieLens, and Jinni. In the past few years, the incredible growth of Web 2.0 web sites and applications constitute new challenges for Traditional recommender systems. Traditional recommender systems always ignore social interaction among users. But in our real life, when we are asking our friends or looking opinions, reviews for recommendations of Mobile or heart touching music, movies, electronic gadgets, restaurant, book, games, software Apps, we are actually using social information for recommendations. In this paper social popularity factor are incorporated in SVD++ factorization method as implicit feedback to improve accuracy and scalability of recommendations. Keywords CF Based Recommendation; SVD; Social Popularity 1. INTRODUCTION Recommender systems have been used for by various ecommerce websites for recommending product, item, movies, music etc. Recommender systems are essential tool in e-commerce on the Web [1]. Nowadays, they are being used by the lot of customer data in existing commercial databases, and more they are available at social networking websites that are most successful system for recommending is collaborative filtering based approach. In order to generate recommendations, Collaborative filtering systems need to compare basically different objects like items against users. There are two main approaches to help such a comparison that make the two main parts of Collaborative filtering approach: the neighborhood approach and latent factor models. Neighborhood methods are based on computing the relationships between users or items. Singular Value Decomposition (SVD) approach, latent factor models, which converting both users and items to the same latent factor that are comparable to each other. The incredible growth of customers and products due to social web and e-commerce websites creates two key challenges for recommender systems. The first challenge is that how to improve the quality of the recommendations for the customers. If quality of the recommender system is good, then customers can trust a recommender; purchase a product, like to book a movies show and finds out he does not like the product, the customer will be unlikely to use the recommender system again. Second challenge is that how to improve scalability of the collaborative filtering algorithms. In somehow there are conflicts in these two challenges. If algorithm spends less time for searching a neighbors, it will be more scalable and worse its quality. We need to consider these two challenges simultaneously so the solutions discovered are practical. We need a new technology that can be useful and dramatically enhance the scalability of recommender systems. Many researchers have suggested that Singular Value Decomposition (SVD) [2] may be such a technology in most of the cases. SVD-based approach can generate results that were much better than a traditional collaborative filtering algorithm most of the time when tested on a MovieLens dataset. But there are some serious limitations when we apply SVD-based approach for recommending which make its less suitable for large scale deployment in e-commerce system. The matrix factorization step is computationally very expensive because it takes a lot of memory and time to factorize a matrix. This is a major problem towards achieving a high scalability while producing good predictive accuracy. In real world e-commerce application, a large number of customers only buy or rate a very small percentage of products, which is real problem. Dimensionality reduction in recommender system is used due to these two problems. That will help to improve the precision of recommendations and reduce the complexity of real time computations. The paper is organized as follows. The next section discusses dimensionality reduction algorithm. Section 3 explains about social popularity concept with integrated model. Section 4 presents experimental evaluation procedure and in Section 5 results and discussion. The conclusion and future research work are given in the last sections. 2. RELATED WORK The aim of Collaborative filtering based approach is to recommend new products or to predict the value of a certain product for a particular user, based on the user’s previous liking and the opinions of same type of users. Several successful systems have been implemented in various organizations like Amazon’s, Netflix, and Last.fm. The weakness of Collaborative filtering based approach for large, sparse databases motivated us to investigate alternative recommender system algorithms. Latent Semantic Indexing (LSI) [3] is used to reduce the dimensionality of user-item ratings matrix. In information retrieval system, LSI is used to solve the problem of synonymy and polysemy. LSI uses singular value decomposition (SVD) as its fundamental dimensionality reduction algorithm, maps well into the collaborative filtering based recommender system challenges. However, SVD cannot be applied to explicit rating in the collaborative filtering based approach because user does not rate most of product so user-item rating matrix have lot of missing values. Furthermore, only few known entries may causes of overfitting [4]. Recent work says that we can fills in missing ratings values and make user-item rating matrix dense. But it is more expensive as compared to other method. Therefore, more recent works recommended that modeling with only the observed ratings, while avoiding over fitting through sufficient regularized model.