Prediction Of Microblog Forwarding Behavior Based On Multi-Feature Fusion

With the rapid development of the Internet in the world,social networks become more and more popular.Social networks connect many individuals in a short time and develop new relationships between people.Through the interaction of social networks,a large amount of data is exchanged and ideas and knowledge are shared.At the same time,the speed of information dissemination on the Internet is very fast.The remarks made by some users on social platforms are irrational.These remarks are likely to be quickly spread into explosive points,develop into harmful public opinion,and cause psychological panic among the social masses.Studying the law and mechanism of information dissemination is of great significance to predict the trend of public opinion and business precision marketing.How to mine effective information from massive data,predict user behavior and improve the accuracy of prediction has become the focus of current research,and it is also the main research problem of this dissertation.This dissertation takes the microblog platform as the research object,and forwarding is the main way of microblog information dissemination.Therefore,predicting the forwarding behavior of microblog users has a certain reference significance for the study of information dissemination mechanism.Among them,the analysis of the influencing factors of user forwarding is the key to improve the accuracy.In the past,the analysis of the influencing factors of forwarding mostly focused on the characteristics of users' personal characteristics and microblog content.The dimension is single.As a weakly connected social network,microblog can spread information and express opinions without constraints,and the emotional information carried can not be ignored,It is also one of the important factors affecting users' forwarding behavior.At the same time,the quality of information extraction also has a direct impact on the accuracy of prediction.We should integrate as much effective information in the data into the prediction model as possible.In addition,the selection of prediction algorithm should also be combined with the specific situation of data to select the appropriate algorithm to achieve the best prediction purpose.To sum up,the research content of this dissertation is mainly divided into the following three parts:(1)Analyze the influencing factors of microblog forwarding.Firstly,this dissertation analyzes and summarizes the research status of influencing factors of microblog users' forwarding behaviors at home and abroad,and analyzes the influencing factors of microblog users' forwarding behaviors from the static characteristics of users and microblog content characteristics.In this dissertation,Python was used to crawl microblog data,including users' personal data,microblog content data,users' following relation data and microblog forwarding relation data,and the data set used in this dissertation was obtained by associating these data.The three dimensions of user behavior characteristics,microblog content characteristics and emotional characteristics are extracted as the influencing factors of user forwarding behavior in this dissertation,which enriches the dimension of analyzing the influencing factors of microblog user forwarding behavior.(2)Extracting feature information.This dissertation selects seven features from three dimensions of user behavior characteristics,microblog content characteristics and emotional characteristics for information extraction.Instead of extracting information from users' static features,user features focus on the concern relationship between users,the number of micro-blogs and the activity of forwarding.The semantic similarity and data form are selected for the content features of microblog.A sentence-bert method based on semantic similarity feature extraction is proposed to obtain a more accurate Sentence vector,and cosine similarity is used to calculate the text similarity between sentences.The characteristics of emotional polarity are subdivided into emotional polarity and emotional intensity.Snow NLP package is used to extract emotional polarity in the microblog text,and the number of emotional symbols is used as the measurement basis of emotional intensity.(3)Build a prediction model for microblog forwarding.Ada Boost algorithm was selected to construct the prediction model in this dissertation,and several S4 vms with low prediction accuracy were combined to obtain a high-precision prediction model.Taking accuracy,F value,accuracy and recall rate as evaluation indexes,the microblog forwarding prediction model proposed in this dissertation is compared with naive Bayes,random forest,logistic regression and support vector machine to verify the validity of the model proposed in this dissertation.In order to further verify the importance of each feature,this dissertation selects the method of grid search for hyperparameter selection,and obtains the importance of each feature in the process of microblog forwarding prediction according to the prediction results.To sum up,this dissertation enriched the dimension of analyzing influencing factors of microblog forwarding and improved the quality of feature extraction.Among them,the semantic similarity analysis method based on Sentance-Bert was used to propose a high-precision Ada Boost prediction algorithm combining multiple S4 VMs,and finally the microblog forwarding prediction model of this dissertation was obtained.It can improve the accuracy of prediction and provide a new idea for the study of communication mechanism in social networks.