The Study Of Behavior Analysis And Application Based On Microblog User

Microblog plays an important role in the social network,and it's also a platform to acquire the latest information quickly for people.the timeliness of microblog is popular with the public.On the one hand,provide entertainment,lifestyle services,information sharing and exchange for the public.On the other hand,achieve the fission of the message.As the most popular social platform,sina weibo has been located in the fore front of Alexa.The study of sina weibo historical data can locate the focus of public concern and look for hot words,which can help government understand the public's value orientation indirectly based on user historical comments at the same time.At present,there are few studies on the three major behaviors based on microblog including the number of forwarding,commentary and praise at home and abroad.Therefore,this article focuses on how to establish the overall forecasting model of the three major behaviors.The specific research contents are as follows:First,this article selects the model features from three aspects: the content of microblog,the level of users and the time distribution.First of all,select text keywords after microblog segmentation.Secondly,the distribution characteristics of the number of different users' forwarding is used to fuzzily cluster which describe the user's historical level.Next,if a microblog is posted when the user is most likely to log in to the microblog platform,the possibility of this microblog being readed will be greatly enhanced.It's important to calculate the average number of forwarding,commentary and praise according to the distribution of time.Finally,this paper builds three model to predict the overall user behavior based on the above features,which is the ordinary KNN algorithm model,the distance-weighted KNN model and FCM-weighted KNN model.Second,this paper selects the overall score as a standard to measure the effect of three model prediction.we also consider the effect of the number of text feature words on the prediction of the model and observe the changes of overall score among these three models.In this paper,100,150,200,250 and 300 text feature words are selected as the input of predicting model.Finally,by comparing these three algorithms,we conclude that when the number of textual feature words is 100,the FCM-weighted KNN model can achieve an overall score of26.62% and is higher than the ordinary KNN algorithm and the distance-weighted KNN model.It shows that the FCM-weighted KNN model can fully mine the microblog information within the experimental range,and this phenomenon also explains that the FCM-weighted KNN model has the common advantages of the ordinary KNN algorithm and the distance-weightedKNN algorithm.