Text sentiment analysis method based on hybrid supervision model

Abstract

The invention relates to a natural language analysis technology, and aims to provide a text sentiment analysis method based on a hybrid supervision model. The method comprises the following steps: performing strong supervision qualitative analysis by using a qualitative sentiment analysis model based on a composite neural network, constructing the composite neural network by combining LSTM and CNN, and extracting sequence features and multi-dimensional features of a text at the same time to more accurately predict the sentiment polarity credibility of the text; achieving weak supervision quantitative analysis based on a syntactic analysis tree, and obtaining the hierarchical modification relation of sentences by conducting word segmentation on the sentences and constructing the syntactic analysis tree; performing recursive upward labeling and calculation according to the sentiment dictionary, and calculating the sentiment intensity value of each sentence; multiplying the credibility bythe emotion intensity to obtain the final emotion intensity of the text. According to the hybrid supervision model provided by the invention, the advantages of two calculation modes in the prior artcan be taken, and an analysis result with credibility and fineness can be given.

Description

technical field [0001] The invention relates to natural language analysis technology, in particular to a text sentiment analysis method based on a mixed supervision model. Background technique [0002] Text sentiment analysis refers to the technology of researching and analyzing the subjective emotional factors in the target text by using related methods in the field of Natural Language Processing (NLP). Generally speaking, the purpose of sentiment analysis is to analyze and judge the emotional tendency or emotional category, opinion, etc. expressed by the author in a given text. [0003] Various existing sentiment analysis schemes can be divided into the following two categories according to the type of training set labels and the granularity of analysis results: Qualitative sentiment analysis gives qualitative sentiment polarity direction and corresponding positive polarity probability value to the analyzed text . The label of its training set has only two possible value...

AI Technical Summary

Problems solved by technology

There are many research results in the field of text sentiment analysis at home and abroad. Support vector machine SVM, naive Bayesian method, maximum entropy model, LSTM, CNN and other models have been applied, but these programs cannot provide reliable text quantitative emotional strength v

Images