International Journal of Computer Applications (0975 – 8887) Volume 113 – No. 2, March 2015 36 Graph based Recommendation System in Social Networks Honey Jindal Department of Computer Science and Engineering JIIT, India Anjali Department of Computer Science and Engineering JIIT, India ABSTRACT Media content recommendation is a popular trend now days. Twitter, Facebook, and Google+ are very popular in the world. The growth of social networks has made recommendation systems one of the intensively studied research area in the last decades. Recommendation systems can be based on content filtering, collaborative filtering or both. In this paper, we propose a novel approach for media content recommendation based on collaborative filtering. Firstly the user-user social network is created using most prominent neighbor set of each user by utilizing their preference information. Then these users are clustered using their neighbor sets and the user with maximum neighbor count is chosen as cluster head. When new user searches for its cluster its similarity is calculated with all the cluster heads. The user gets recommendation based on the average ratings of his cluster members. The proposed approach is tested on the users of Movielens Dataset. The proposed approach gives a hit ratio of 89.33%, Mean Absolute Error as 0.4756 and Root Mean Square Error as 0.7671 on Movielens dataset. Keywords Recommendation, social networks, content filtering, collaborative filtering, clustering, preferences, neighbor set. 1. INTRODUCTION Recommendation system plays an important role in our lives. Accurate recommendations may help user to quickly identify their desirable items. It is widely used in online commercial sites to satisfy user’s personal demands on the basis of their purchase behavior. Recommendation systems automatically recommends items to the target users based on their past purchases and behavior [1]. For example, a customer usually expresses an interest in an item either by viewing a product description or by placing the item in his “Shopping cart”. Thus the customer will likely receive recommendations for additional products. These products can be recommended based on the top overall sellers on a site or on an analysis of the past buying behavior of the customer as a prediction of future buying behavior. Items are recommended to users based on their past ratings. Recommendation systems can be summarized in three techniques: Content based filtering, collaborative filtering and Hybrid filtering technique. Content based filtering recommends resources to user according to past purchase history of the user. A major disadvantage of this method is that it uses only static information and the user interaction information is not fully utilized. Collaborative filtering relies on the past preferences or ratings correlation with other users. Based on this correlation, people with similar preferences are taken into account for recommendation. Hybrid methods are the combination of both content based and collaborative filtering. The popular online social networking websites such as Facebook, twitter and YouTube provide novel ways for people to communicate and build virtual communities. The online shopping websites such as Amazon [1] [3], Flipkart and Snapdeal provides recommendation based on the purchase history and recently viewed items. This paper proposes a novel approach for media content recommendation in social network based on collaborative filtering. Firstly, a similarity matrix is calculated based on Euclidian distance between each user pair. Afterwards, a social network is formed where node represents the user and edges represents connection among user and the prominent neighbors with high similarity score. Then the clusters are formed based on the social network by k-means clustering algorithm. The node with highest closeness centrality is selected as the cluster head. When a new user arrives, its preferences are compared with the preferences of all cluster heads. The most similar cluster head is allotted to the new user. Thus the new user belongs to the most similar cluster and the average ratings of the cluster are recommended to the new user. This approach reduces the online computation time in cluster determination for the target user. The remainder of this paper is organized as follows. First, we summarized the related work in section 2. Section 3 illustrates the proposed method in detail. Section 4 includes complexity analysis. Section 5 shows the result and simulation of proposed method. Section 6 gives discussion, followed by conclusion and future directions for research in section 7. 2. RELATED WORK In this section, some of the prior techniques for recommending resources to the users are briefly reviewed. A recommender system is an effective tool used to reduce information overload while searching content, product information or documents on the internet. The recommendation problem occurs each time as user enters in the system. Consider a recommendation system with M users and N items shown in Fig.1. The relationship between user and item is represented by M*N rating matrix, R. Each entry in R ij represents rating given by the user i to item j which scales from 1 to 5. An item with no rating represents as R ij =0. Recommendation algorithm is the core of every recommendation system, which determines the performance of the system. There are three popular methods adopted by most of the recommendation systems. Firstly, the content- based recommendation method [4] [7] [8], which recommends resources based on the past purchase history and not on user’s preferences and opinion. One of the disadvantages of this method is that each resource is defined by their associated features and resources are recommended to the target user based on similarity of the resource features not by utilizing the taste and preferences of the target user.