Research And Implementation Of Friend Recommendation Based On User's Relationship Intensity

Social Network Services (SNS), which is one of the most popular applications, is increasingly altering our dating patterns and interactive way with its innovative, convenient and practical features. With the rapid growth of subscribers, it is becoming more and more important and difficult issues for ordinary users to expand their circles and find friends from the complex web of relationships. In this context, various forms of "Friend recommendation algorithm" come into being.In calculating the "customer relationship strength", the view of this paper is to split it into two phases, namely, "Interest-based communities found in topic similarity algorithm" and "Establishment of the comprehensive similarity model based on link information. " These two stages exist in strict order that taking the results of the first phase of the output as the input to the second stage.The first stage of the algorithm is "interest-based similarity algorithm of community discovery", whose main work is to calculate the "user-interest similarity", and to divide this community according to the similarity. To this end, this paper introduced LDA topic model to extract hidden topics of interest in social networking sites, accessing all users'probability distribution of each theme, and gather users in a "interest community " by the "edge cluster" approach.The second stage is the "Establishment of the comprehensive similarity model based on link information", whose main work is to quantify the relationship between users'link. In this process, trust models and local random walk model were constructed. Wherein the trust model is mainly considered the number of common friends between the two users, and regard it as an initial value of the trust, and expand the model by spreading the theory of trust; local random walk model borrowed the idea of PageRank, focusing on the information of the out-degree and in-degree of the user node. The final recommendation was given by integrating the two models.In accordance with the above ideas, this paper uses Java language to realize the friend recommendation algorithm. In order to verify the feasibility and effectiveness of the algorithm, this paper is to compare "Friend recommendation algorithm based on user's relationship intensity" with "Friend recommendation algorithm based on content", and evaluate the proposed algorithm from the accuracy rate, recall rate and F value.Experimental results show that the "Friend recommendation algorithm based on user's relationship intensity" achieved a better recommendation quality compared to the other methods, not only improved the recommendation accuracy, and effectively reduced the algorithm"s time and space overhead.