Social text sentiment classification method and system based on text graph neural network

Abstract

The invention discloses a social text sentiment classification method and system based on a text graph neural network, and belongs to the technical field of natural language processing. Comprising the following steps: receiving a target text, and removing an abnormal value in the received text; obtaining word embedding of the target text by utilizing a BERT model; obtaining emotional polarity characteristics of the target text, calculating an emotional score of each word of the target text by using a SentiWordnet emotional dictionary source, and taking a final score of each word as the emotional polarity characteristics of the word; splicing the word embedding and sentiment polarity features of the target text to form an initial word vector; and constructing the target text into a text graph structure, taking the initial word vector as a node initial feature of a text graph, then performing feature extraction by using a text graph neural network message passing mechanism, and finally performing sentiment classification on the extracted feature. The method not only considers the context features in the speech, but also considers the mutual relation between the speech, so that the sentiment classification is more accurate.

Description

technical field [0001] The present invention relates to natural language processing technology, in particular to a social text sentiment classification method and system based on a text graph neural network. Background technique [0002] Regarding the sentiment analysis of social media speech, a lot of research and exploration have been carried out at home and abroad. At present, the methods for social media sentiment classification can be divided into two categories: sentiment analysis based on semantic dictionary and sentiment classification based on machine learning. [0003] Emotional words refer to words with emotional tendencies. The identification methods of emotional words in social media can be divided into methods using dictionaries to calculate similarity, and statistical methods based on large-scale corpora. [0004] The discrimination method based on the sentiment dictionary is to use the dictionary to directly judge the polarity of the corresponding words, and...

AI Technical Summary

Problems solved by technology

The advantage of the deep learning model is high accuracy, but there are also some obvious disadvantages, such as time-consuming training, and the inability to explain the semantics of the final decision, etc.

[0007] For example, the input of the deep neural network is independent speech, and the contextual features between long-distance words are fused together through the attention mechanism, but the correlation features between speeches are ignored, and this type of deep learning method cannot be interpreted

Images