User Behavior Analysis Based On Social Network Data

With the rapid development of technology,human beings have entered the era of mobile Internet.According to data released by KPCB in 2017,as of the end of 2016,the number of Internet users exceeded 3.4 billion,and the penetration rate was as high as 46%.The number of online social networking platforms such as Facebook,Twitter,WeChat,Sina Weibo,etc.continues to grow,and the number of users is huge.The profile of one person can be reflected by his/her posts online.By analyzing the information hiding by these posts,we could learn one's behavior pattern and give a reasonable prediction of one's future behavior.Recently,how to analyze online social network and how these analysis can benefit the real life has been a researching hotspot for researchers at universities,research institutes,and companies.In this paper,we focus on mobile social network data from individuals and perform a time series analysis of the behavior patterns extracted from it.The research process of this paper is as follows: Firstly,we select Sina Weibo,which is the largest micro blog service in china,as the data source.And design an efficient crawler to crawl a large number of user public data from Sina Weibo.Secondly,due to the large number of users,we need to select valuable users to analyze,we calculate user weight using PageRank,and the target user is selected according to the weight calculated.Thirdly,we process a large amount of microblog text data that is crawled,and use the TF-IDF algorithm to process the text data after the word segmentation,extract the keywords as user post representative,and classify the post into corresponded fields for future analysis.In the fourth step,this paper proposes a microblog theme probability prediction model based on Seq2 Seq.At the same time,through the actual experience analysis,we propose a hypothesis that the behavior of user when using weibo is time related.Based on the hypothesis,an improved model of the Weibo theme probability prediction based on Seq2 Seq and time feature is proposed.In the fifth step,the models proposed in this paper are compared with the popular prediction model.For the presentation of research results,this paper first presents some statistical information of crawled data.Then there is a visual representation of the characteristics of the selected target user.Finally,the proposed Seq2Seq-based Weibo topic probability prediction model and its improved model are compared with the popular time series prediction model ARIMA model and Holt-Winters model.In this paper,the prediction effects of different models are compared by experiments.The results show that the proposed model is better than the ARIMA model and the Holt-Winters model in short-term prediction.At the same time,the improved model gain much improvements in some topic.This shows that the proposed model has better prediction accuracy compare to the traditional statistical prediction model in terms of short-term prediction of the Weibo theme probability.It also proved the hypothesis proposed.This result has a good reference for Weibo promotion and marketing.