Research On Recommendation Methods For Network Users Based On Social Influence

Recommender systems provide suggestions for network users to select commodities, users can make right decisions in selecting commodities and the service providers can also improve the service quality and increase profits by making use of the recommender systems. In this paper, we study the recommendation methods for network users based on social influence. This paper builds a commodity ranking recommender systems simulation model through the analysis of the behavior characteristics of network users in listening and downloading songs. Then simulation experiments show that it can work well and find the impact of commodity ranking recommender systems. In order to improve the prediction accuracy of recommendation methods, our approach, which considers the impact of social influence on the basis of the traditional collaborative filtering recommendation methods, proposes collaborative filtering algorithm based on trust and tags. Finally, the analysis method based on tags and commodity ranking is applied to the data set sourced from Last. fm music web site, we further discuss the influence of tags and commodity ranking to the network market. The main work is as follows:(1) This paper in-depth analyses the behavioral characteristics of listening songs and downloading songs in recommender systems on the basis of experimental study results, and proposes a two-stage simulation model of commodity ranking recommender systems. In the first stage, network users decide which songs to listen only affected by the social influence. In the second stage, network users decide which songs to download only affected by the quality of songs. Experimental results show that the simulation is excellent in fitting and finding the impact of commodity ranking recommender systems. Lastly, we give reasonable recommendations of avoiding the malicious attacks in recommender systems.(2) This paper proposes collaborative filtering algorithm based on trust and tags. Our approach, which joins the impact of social influence of trust and tags on the basis of the traditional collaborative filtering recommendation methods, selects the recommender users in accordance with user’s tags preference to provide recommendations. According to the number of co-occurrence tags finds strongly correlated tags set. The algorithm clusters the tags according to the results of strongly correlated tags set, and it makes the relations between the clusters are low. At last, we generate user recommender weight matrix and give rating predictions based on collaborative filtering algorithm involved in trust and tags. (3) Based on the commodity ranking recommender systems simulation model and the collaborative filtering algorithm based on trust and tags, the analysis method are applied to the data set sourced from Last, fin music web site. First of all, we make the statistical analysis of the user’s tags according to clustering results of tags, and we have the clusters of users tag preferences. Then, we calculate songs’popularity and songs’tag attributes. Lastly, we rank the songs’popularity in different users’clusters and the whole network users to analysis the influence of songs ranking to the songs’market. The experimental results show that the inequality of success in the songs’ market, and the different songs’preference among different users’clusters.