Research on the Topic Content Prediction of Online Reviews: from the Perspective of Bi-directional Sentiment Classification

Abstract

To solve the problem of coarse-grained results in the research of topic content prediction, this paper proposes a prediction method for the topic content from the perspective of bi-directional sentiment classification. Firstly, the method uses SnowNLP to classify the sentiment of online reviews; secondly, LDA model is applied to extract the topics and entropy is used to sort topics; finally, Word2Vec is applied to achieve the prediction of the topic content. Example calculation shows that this method effectively solves the problem of coarse-grained prediction results of online reviews’ topic content, and presents the prediction results from positive and negative sentiments. The average precision of positive topics is 86.67%, and the average precision of negative topics is 80.00%.