A Cross-Domain Recommendation Algorithm Based On Social Network

Recommendation system is considered to be an effective solution to information overload. Personal recommending service has achieved great success in e-commerce. Besides, it also shows its significance in many other areas, like social networks, academic cooperation, information technology etc. Though, there still exist two problems for traditional recommendation system, they are data sparseness and "cold start".Cross-domain recommendation system is proven to be a solution to the two problems. It aims to summarize the knowledge of the auxiliary domain, then transfer the knowledge to the target domain based on special relation or algorithms. In this paper, we proposed a cross-domain recommendation algorithm in which we utilize the friends network to find the similar users.First, we describe the basic idea of the algorithm in the paper. We pick up TopN similar users from the friends network of the target user. Then we propose an algorithm to calculate the similarity between users. Finally, we can predict the item ratings in target domain based on the similar users’data. In the whole process, the key lies in the algorithm of link prediction in friends network. To get a better performance, we apply the PageRank random walk in friends network. Furthermore, we propose four factors to optimize the link prediction, they are common friends number, domain number, common domain number and enhancing cross links. We explain the basic theory and impact of each factor in detail in this paper, and present the algorithm of cross-domain recommendation.What’s more, we conduct the evaluation on Yelp datasets, presenting the influence on the process of random walk for each factor. Finally we proven the superiority of the algorithm proposed in this paper.