Research On The User Classification Algorithm Of Online Social Network

In recent years,with the popularity of the Internet,all kinds of online social networks and online services arise at the historic moment,social networks have become the main platform of people daily life communication.Social network takes users as the main body and the interpersonal relationships in real life as the prototype.With the help of the platform the users interact with others and the disseminate information.Social network in bring convenience to people at the same time,also has brought all sorts of problems.Because of frequent online interaction and the explosion of information,on the one hand,the users are unable to choose the valuable information,on the other hand,the attacker spreads false and malicious messages by posing as normal users,threatening users' privacy and the right values received threats.So how to extract useful information from vast amounts of information and make personalized recommendation and detect abnormal users becomes key problems.Aiming at this problem,according to different requirements,this paper takes Sina microblogs as the research object to user classification and the key research problem is to identify different users water-forces in online social networks.the main research is as follows:1.Based on the lack of existing mutual information algorithm,an improved feature evaluation function is proposed.In traditional mutual information,due to ignore word frequency factors,the function of low-frequency feature is amplified and the useful feature is omitted.Aiming at this problem,the characteristics of the evaluation function introduces weighting factor,the intra-class factor and the inter-class factor to make up for its shortcomings.And the experimental results show that the improved method is better than the existing method of mutual information in performance.2.On the base of attribute reduction,in view of the existing weighted Naive Bayesian algorithm,it is not possible to take the limitation of global consideration and the random blindness of the initial weight value when obtaining the weight value,so the particle swarm optimization algorithm is adopted to optimize the weight value by using the word frequency ratio as the initial weight value,and then the weighted Naive Bayesian classifier is constructed.And the experimental results show that the improved the Naive Bayesian algorithm is superior to the existing Naive Bayesian classification algorithm.3.On the base of attribute reduction,in view of the existing support vector machine algorithm,it is only the problem of unilateral improvement in kernel function and multiple classification model.Aiming at this problem,first of all,the particle swarm optimization algorithm is used to iterate the parameter combination of the mixed kernel function and then a multi-classification support vector machine classifier is constructed by combining improved decision tree multi-classification model.And the experimental results show that the improved multiple classification support vector machine algorithm in classification precision is superior to the existing support vector machine classification algorithm.