Research Of Social Network User Interaction Model And Behavior Preference Prediction

The development of Internet and mobile communication technology brings innovative boom to businesses, and as the penetration of user centric concept, analysis and understanding of user behavior has become a significant means to enhance the user experience. Further, the popularity of all kinds of social networks and intelligent terminals, it provides massive and real data for the analysis and prediction of user behavior and preference.However, there are still some problems to be solved on the analysis and prediction of user behavior based on social network. First of all, in the model building of social user interaction network, it includes two issues:the generation of user social relationships and the measure of user preference consistency. On the one hand, with the restrictions of user themselves and other factors, the social relationships of most social users are incomplete and sparse, and it restricts the coverage and accuracy of user behavioral prediction. On the other hand, user often builds social relationship based on their interests on social networks, and that the preference for consistency between users is also different. It is very important to measure the differences of user preference for improving the precision of user behavior prediction. Second, as an important attribution of social networks, social influence has been applied in user behavior prediction model. On social networks, user behaviors, ideas, options, thoughts are often affected by their social relations, and the social influence will propagate along with user links, such as famous "three degree influence theory". It is the key of social influence based prediction model how to measure user social influence and calculate the influence propagation on social networks. Moreover, it is also a valuable research how to evaluate the results of user behavior preference prediction model from the macro and microstructure.In response to build user interaction network, user prediction model and evaluation model, the paper will focus on the research of user behavior preference prediction. With the aid of user graph similarity, it extracts user potential social relationship. And by making user of user preference information, it designs effective user preference consistency algorithm. In user behavior preference model aspect, the local and diverse user social influence prediction model will be researched. Meanwhile, it proposes a visualizable evaluation method, which also can evaluate the performance of prediction algorithm from micro level. The main research contents and major contributions of this paper are as follows:Due to the features of social networks, the limitations of user’s time and energy and other factors, the user social relationships are often incomplete. For most users, their social relationships are very sparse, which will lead to the incompletion of user social network and it is not conductive to user behavior analysis and prediction. Node similarity based algorithm is one of the most simple and popular user hidden relationship mining methods. However, its prediction accuracy also needs to be further improved. Considering the significant impact of weak ties to the likelihood of user connection, the paper proposes a user potential link mining algorithm which is based on weak ties and node centrality and it is effective to improve the precision of extracting hidden relationships.In social networks, as the bond of the social circle’s forming and maintaining, user interest is an important indicator to measure user relationship. However, the calculation of existing user similarity methods has some shortcoming, such as low precision, low discrimination, which can not represent the preference similarity between users. Hence, on the basis of existing methods the paper proposes a new novel heuristic user similarity computing model, which considers the micro and macro affecting factors at the same time. Further, the model improves the precision of user similarity, and provides a high distinguishing ability between those very similar users.The social influence and propagation is one of important attributions on social networks, which attracts widely recognized and research interests by researchers. The user’s behavior, ideas, decisions are often affected by their social friends, and the social influence will be propagated along with the social relationships. Through the understanding of social influence and propagation between social users, we can analyze and predict the trend of user behavior. In the calculation of social influence, the existing methods either lack the social influence of global diversity, or need to know the whole network information. Hence, the paper proposes a computation method of social influence, which is based on local node network topology and user local interaction respectively. The proposed calculation of social influence will be restricted the range of neighbors. The extracted social influence is local, diverse and low computational complexity. For the influence propagation, a shortest and maximum influence propagation path strategy is adopted.The effective evaluation model can help choose the most suitable prediction algorithm under different scenarios. The most existing evaluation criteria are established on the theoretical basis of Bell distribution, which exist some shortcomings, such as the fine grain size is not enough, it can not adjust with the changes of user experience, etc. In this paper, we find that many prediction results are not meeting the Bell distribution, but approximately obey the power law distribution. Hence, based on the cumulative probability distribution model, a visualized evaluation method is introduced. According to this model, we can compare different prediction algorithms from the more granular level. Meanwhile, the evaluation expectations can be calculated according to the cumulative probability distribution. So, we can discrete the predictive results according to the particle size of user experience. Hence, the evaluation results will better suit the user experience and scenes.