Weakly-supervised object recognition method for case-related microblog evaluation based on variational dual-topic representation

Abstract

The invention relates to a weakly supervised case-involved microblog evaluation object recognition method based on variational dual-topic characterization. The present invention first uses the neural topic representation network to perform two variational coding and reconstruction on the comments to obtain the topic features; then, calculates the reconstruction loss of the neural topic representation network, and then uses the label samples to calculate the classification loss of the evaluation object; finally adopts joint training The strategy is to jointly optimize the reconstruction loss of the neural topic representation network and the evaluation object classification loss to realize the automatic identification of Weibo evaluation objects. The present invention first uses the variational dual topic representation network to encode and reconstruct comments twice to obtain rich topic features, and then uses a small number of tag comments to guide the topic representation network to automatically identify and evaluate object categories. Automatic classification of evaluation objects and mining of evaluation object terms are realized. The theoretical and technical verification was carried out on two datasets of public opinion involved in the case, and the experimental results fully proved the effectiveness of the method.

Description

technical field [0001] The invention relates to a weakly supervised microblog evaluation object recognition method involved in a case based on variational dual-topic representation, and belongs to the technical field of natural language processing. Background technique [0002] Negative emergencies related to the case usually trigger heated discussions among netizens on Weibo on the Internet, and form a hot topic that spreads quickly and has a wide range in a short period of time, thereby generating public opinion on the Internet about the case. Identifying the evaluation objects concerned by the public opinion involved in the case from a large amount of comment corpus, such as legal institutions, parties, media, etc., is the basis for tasks such as public opinion analysis and situation assessment. The specific task of the microblog evaluation object recognition involved in the case is to identify the object terms from the comment corpus, gather the evaluation object terms w...

AI Technical Summary

Problems solved by technology

However, the use of this type of neural topic model in the identification of Weibo evaluation objects involved in the case still has the following shortcomings: (1) The model only reconstructs the text once, which limits the model's learning of topic representation

Images