Microblog Popularity Prediction Based on Multimodal Feature Fusion

The change of public opinion popularity shows complex non-linear characteristics under the joint action of many influencing factors. Taking microblog as the research object, considering the real-time and complexity of microblog, this paper analyzes the trend of microblog popularity from the aspects of microblog content, user characteristics and microblog event timeliness, and proposes a multi-modal feature fusion microblog popularity prediction algorithm. Firstly, the original features of microblog are analyzed, key features are extracted from them, and GDBT algorithm is used for feature screening. Secondly, the popularity of microblog is measured according to the information gain method. Finally, combined with XGBoost algorithm, the popularity of short-term events is predicted according to the text content, image content, video content and user influence of microblog, thus predicting the popularity value of users whose microblog events have been published, and then using LSTM algorithm to predict the overall event trend. Experiments show that. The accuracy of XGBoost algorithm in predicting the popularity value of microblog can reach 80%, while LSTM algorithm has a small RMS for predicting the overall event trend, which well predicts the overall development trend of events.