Research Of Social Distance-aware Bayesian Personalized Ranking Recommendation Algorithm

Recommendation system recommends new unused items to users according to their preferences which is learned from the historical feedback of user.There are usually two forms of user's historical feedback which are explicit feedback and implicit feedback in recommendation system.In general,explicit feedback is the numerical rating of the item by the user and implicit feedback is the user's binary description of the behavior to item.In many practical applications,explicit feedback is hard to obtain,so the recommendation system needs to use implicit feedback which is easy to obtain and enormous to learn users' preferences.Because explicit feedback is very sparse,rating-based method to estimate the numerical score of item by the user cannot perform well.Ranking-based method which obtain the ranking list of the item by estimating the user's preference can avoid the sparsity of implicit feedback.Recent years,social network has been incorporated to recommendation system to improve the accuracy of recommendation.Because friends in social networks usually have similar interest,users' own preferences can be estimated based on their friends' preferences.Existing social recommendation methods only consider the users' direct friends in the social network,but fail to consider the propagation process of the influence between users in the social network.Existing social recommendation methods fail to make full use of the graph structure information of the social network,thus limiting the accuracy of recommendation.Aiming at the weakness,the SDBPR(Social Distance-aware Bayesian Personalized Recommendation)method is proposed.The core idea is to avoid the sparsity of implicit feedback data by directly modeling the ranking of users' preferences and to make full use of the structure of social network by considering the friends of multistep distance to simulate the propagation process of user influence in the social network.Specifically,SDBPR first uses the random walk algorithm to travel and sample the social network to obtain the random path,and generates the binary ranking hypothesis of users to items based on the distance between users on the random path.Then the probability of these binary ranking hypotheses is calculated by bayesian method,and the user and item feature representation is obtained by maximizing the probability through the stochastic gradient descent method.Finally,the preference value of the user to the item is calculated by using the feature representation of the user and the item.,and then the preference ranking list of the user to the item is generated.In the experiment,SDBPR model was tested on two real data sets and compared with other existing methods.The experimental results show that the recommendation performace of SDBPR model outperform other methods.