Text viewpoint mining method and system

Abstract

The invention relates to a text viewpoint mining method and system. The method comprises the steps of obtaining a to-be-mined text; carrying out preliminary screening of viewpoint statements on the to-be-mined text; according to the screened text to be mined, adopting a Fast Text-XG Boost model to carry out viewpoint statement recognition; according to the recognized viewpoint statement, adoptingan MTL-BiLSTM-CRFs model to perform viewpoint mining with annotation of the evaluation object and the evaluation word as a main task and part-of-speech annotation as an auxiliary task. According to the text viewpoint mining method and system provided by the invention, the mining efficiency and accuracy of the text viewpoints are improved.

Description

technical field [0001] The invention relates to the field of opinion mining, in particular to a text opinion mining method and system. Background technique [0002] At present, this paper has carried out the following two aspects of research on opinion mining methods: 1) research on opinion sentence recognition; 2) research on opinion mining. At present, many researchers at home and abroad have carried out research on the above two aspects. [0003] At present, the methods of opinion sentence recognition are mainly divided into rule-based methods, bootstrapping optimization-based methods, classifier-based methods and graphical model-based methods. [0004] Li Yongyong formulated a five-level emotional dictionary, polarity dictionary and negative thesaurus, and on this basis, he proposed non-opinion sentence discrimination rules and opinion sentence discrimination rules to distinguish non-opinion sentences from opinion sentences. Min Hou constructed a sentiment dictionary b...

AI Technical Summary

Problems solved by technology

[0014] The method of identifying opinion sentences based on rules can quickly distinguish between opinion sentences and non-opinion sentences, but the formulation of rules is limited by the corpus and language itself, and the recognition process is relatively limited in different fields

Moreover, due to the short length of Weibo text, fragmented content and messy sentence structure, if the text is not semantically analyzed and the classifier is directly used to train the extracted features, the accuracy of opinion sentence recognition will be reduced.

Using the BiLSTM-CRFs model can not only capture the forward and backward information of the text, but also ensure the order of label annotations, and solve the problem of scarce words in sentences, but ignores the evaluation of words with different parts of speech in sentences. Objects and evaluation words have different weights, and the parts of speech of evaluation objects and evaluation words are not unique, and are more likely to be a combination of multiple parts of speech, resulting in low efficiency and accuracy of text opinion mining

Images