Research On The Method Of Social Network Link Prediction Incorporating Asymmetric Interaction And Personality Link Preference

With the rapid development of mobile Internet,the penetration rate of social service platforms is getting higher and higher,and the service content has been extended to all aspects of life.Social network is not only an extension of the real society on the network,but also a new social model closely intertwined and parallel and independent with the real society.Social network has become an important way to find friends and expand social circle,and how the platform service providers recommend more suitable friends for users to improve user experience is how to improve the accuracy of link prediction,which is also the main research problem of this dissertation.At the same time,the scholars in the study of personality traits,proved that personality can be used to predict the behavior of users,including academic achievement,health and social networking behavior,etc.,so whether the personality can also affect the links in the social network relation,as link whether personality can be predicted a basis,it is also in this dissertation,we study a problem.In view of the above problem,this article have to the social networks of the link in order to improve the prediction accuracy as the main target,through the social platform first data to predict the user's personality trait score,and the analysis of the effect of personality in the social network links,then consider many factors in asymmetric interactions and personality preferences links prediction model.The main research contents of this dissertation are summarized as follows:(1)training the personality prediction model applicable to microblog platform.First collect seeds by means of weibo small application node id and personality trait score,and then write the crawler crawl seed node and extension of basic information,weibo data,such as the content,the fans and focus on the list of data screening,code,and after the pretreatment,according to the characteristics of the optimized design scheme(including static properties,text,language characteristics and the dynamic behavior)sorting out the feature set,including emotional characteristics in emotional calculation for weibo,expand the network emotion,dictionary as emoticons and combined application of Pearson correlation coefficient and Lasso hybrid feature selection method to select the final feature set,Finally,the personality prediction models of linear regression,random forest and decision tree were trained simultaneously,and the random forest model with the best regression effect was selected for the personality prediction of extended nodes.(2)definition of personality link preference.Users were divided according to the score of five personality dimensions,and the distribution of personality of users with different personality characteristics who paid attention to users was analyzed,and it was found that there was difference in the distribution.This dissertation defined this difference as link preference.In addition,it is observed that there are correlations among the five personality traits,such as openness and extraversion positively correlated and neuroticism negatively correlated,but there are also unrelated traits,such as agreeableness and openness.Therefore,this dissertation considers the link preference of five dimensions when calculating personality link preference.(3)put forward a four-dimensional comprehensive link prediction model.Proposed in this dissertation to make links to social networks to predict,at the same time considering the user attribute similarity and network structure compact connection degree(including tightness and improvement based on common neighbor node based on the tightness of the path),and from different directions the asymmetry between the user interaction strength,then this article links defined preferences are used to calculate based on personality traits link preference matching degree,and then using existing data link to the four dimensions of weight training,to determine the optimal weight after calculating the final link prediction probability.Finally,the effectiveness of the model's connection strength,asymmetric interaction strength and link preference matching degree in improving the accuracy of link prediction was verified through comparative experiments.To sum up,this dissertation optimized the personality model for Chinese social networking platform,to find and define the existence of personality preferences link,put forward a view of links to social network forecast model,proved that the asymmetric interaction strength and personality will affect the social network link relationship between users and ultimately achieve the goal of improve the predictive accuracy links.