A Method of Extracting Attribute Words

Abstract

The invention discloses an attribute word extraction method, which provides a neural network topic model structure with double encoders, uses a preset attribute encoder and an auxiliary encoder to encode a comment document, solves the problem that emotion expression deviation in the comment document cannot be encoded, and introduces weakly supervised seed word information, the learning effect of the double-encoder neural network topic model is improved, the knowledge distillation thought is introduced, the attribute encoder and the auxiliary encoder serve as a teacher model and a student modelof knowledge distillation, combined learning of double encoders is achieved, and iterative training is completed. Compared with the prior art, the method has the advantages that the attribute words in the comment text can be accurately extracted, and the accuracy of sentiment analysis is improved in a fine-grained sentiment analysis task.

Description

technical field [0001] The invention relates to the technical field of natural language processing, in particular to an attribute word extraction method. Background technique [0002] Sentiment analysis is an important task in Natural Language Processing (NLP), and its purpose is to analyze subjective text with emotional color. Sentiment analysis can be divided into three levels: text level, sentence level and attribute level. Among them, attribute-level sentiment analysis is a sentiment analysis task for specific attributes. It can mine users' emotional tendencies at a finer-grained level, so it has become one of the current research hotspots. [0003] Attribute-level sentiment analysis is mainly divided into two steps: 1) attribute word extraction and 2) emotion polarity recognition, the former digs out the evaluation objects involved in it from the comment corpus, that is, some attributes of the product, and the latter judges the text for this Attributes express emotion...

AI Technical Summary

Problems solved by technology

The existing attribute word extraction methods include supervised methods and unsupervised methods. The supervised method requires a large amount of labeled comment corpus for model training, and the domain transferability is poor, which limits the practical value of the method.

The unsupervised method does not need to label data and has good domain transferability, but the disadvantage is that it lacks prior supervision information and the model accuracy is low

Images