Research On Social Behavior Prediction Technology Based On Feature Learning

In recent years,social media has gradually become a very important part of people's life.People are used to express their opinions and ideas on social platforms.For example,on Sina Weibo,users express their opinions and share their interests by publishing Weibo.When users see the interested Weibo,they can express themselves by forwarding Weibo,commenting,praising,etc However,in the era of big data,social platforms have the problem of information overload,and users can see too many microblogs,so it is difficult to find interested microblogs from massive information.In this case,how to predict users' social behavior has become a key issue for researchers to study social platforms.This paper mainly introduces the research work of user forwarding behavior prediction technology based on social platform,in which the prediction method based on user historical behavior is mainly studied.By combining collaborative filtering algorithm and sequence recommendation,two new models are proposed to solve the problem of user transition to distribution prediction.Firstly,the research status of social behavior prediction is introduced;secondly,the user historical behavior is introduced At last,two new models are proposed.The first model is to express the sequence characteristics of microblog forwarding path through the embedded vector of LSTM network learning microblog,which realizes the combination of deep learning based collaborative filtering model and sequence recommendation model NCF-L model,compared with the basic model NCF,has better recommendation effect,and effectively embeds sequence features into collaborative filtering model;secondly,a model is proposed,based on the idea of graphrec model,trying to add social relations into the model,learning the sequence feature vector representation of active neighbor set through attention mechanism,and combining with NCF model,NCF is proposed-Based on NCF model,NCF-A model also improves the recommendation effect;compared with NCF-L,NCF-A model directly uses social network relationship data to add the sequence characteristics of active neighbor sets to the model.In order to prove that the model proposed in this paper has good recommendation effect and generality,the traditional collaborative filtering algorithm and serialization recommendation algorithm based on deep learning are compared with the model proposed in this paper on multiple datasets,in which HR and NDCG are used as the evaluation indicators;through experiments on each model,the recommendation effect of the two models proposed in this paper is fully proved Better,it is not only suitable for the prediction task of forwarding behavior on microblog social platform,but also suitable for other social platforms,which can do a good prediction and recommendation of users' social behavior on social platforms.