Research On Forecasting And Recommendation Model Of Hot Topics On The Internet

With the rapid development of social networks,the relationship between people is gradually networked.Social networks have affected many aspects of people's lives.However,due to the exponential growth of social network data,people have encountered diversity and convenience in social networks,but also encountered the problems of complex choices and information interference.Therefore,this paper conducts research on hotspot prediction and recommendation models for topics in social networks.Internet hot topics not only have the characteristics of real-time and diversity,but also compared with traditional media news,the topic text also contains high dimensionality,sparse data,and irregular language,etc.,which leads to poor classification and recommendation of hot topics.The problem of stability and low accuracy.This article studies and analyzes the problems in the classification and recommendation of hot topics.The main work includes the following aspects:(1)Aiming at the hot topic prediction framework design problem,this paper designs the hot topic prediction structure framework based on the text mining process,and designs the hot topic prediction process framework based on the CRISP-DM standard process.Aiming at the problem of feature extraction within topic text,this paper constructs a hybrid feature vector.This mixed feature vector space takes into account the characteristics of the text content of the hot topic,and also considers the numeric features such as the number of reposts and the number of comments of the hot topic.(2)Aiming at the design issues of hot topic classification models,this paper comparatively analyzes the performance of three classification algorithm models of logistic regression,SVM and random forest.Based on three classification algorithms of logistic regression,SVM and random forest,this paper proposes a weighting-based classification algorithm.Voting combined classification algorithm.The experimental results show that the combined classification algorithm has high accuracy and stability,and the F1-Score reaches 95%,which is higher than that of a single classification algorithm.(3)Aiming at the time factor in topic recommendation,this paper establishes a dynamic hybrid recommendation algorithm model.Based on the mixed recommendation model based on content and collaborative filtering,this model introduces the time factor to increase the characteristics of the user's own real-time changes,and solves the problem that the target user's preference characteristics remain unchanged.The experimental results show that the dynamic hybrid recommendation algorithm model has a hit rate of 85% and an accurate recommendation effect.