Sentiment analysis model pre-training method, device and electronic equipment

Abstract

This application proposes a sentiment analysis model pre-training method, device and electronic equipment, which relate to the technical field of artificial intelligence. Wherein, the method includes: based on a given seed sentiment dictionary, performing emotional knowledge detection on each training corpus in the training corpus, determining the detection emotion words and detection word pairs contained in each training corpus; according to the preset mask processing rules , each training corpus is masked to generate a masked corpus; use the preset encoder and decoder to encode and decode the masked corpus to determine the predicted emotion contained in each training corpus Words and predicted word pairs; according to the difference between the predicted emotional word and the detected emotional word, and the difference between the predicted word pair and the detected word pair, the preset encoder and the preset decoder are updated. Therefore, by using this sentiment analysis model pre-training method, the pre-training model can better represent the data in the direction of sentiment analysis, and improve the effect of sentiment analysis.

Description

technical field [0001] This application relates to the field of computer technology, in particular to the field of artificial intelligence technology, and proposes a method, device and electronic equipment for pre-training an emotion analysis model. Background technique [0002] Sentiment analysis refers to the study of people's opinions, attitudes, and evaluations of products, services, organizations, and other entities. Usually sentiment analysis includes multiple sub-tasks, sentiment analysis, comment mining, entity-level sentiment analysis, sentiment analysis, etc. At present, the sentiment analysis of text can be realized through the sentiment analysis model. [0003] In related technologies, a deep neural network can be used to perform self-supervised learning on large-scale unsupervised data to generate a pre-trained model, and then in a specific sentiment analysis task, based on the emotional labeling data of this type of task, transfer learning to the pre-trained m...

AI Technical Summary

Problems solved by technology

[0005] The sentiment analysis model pre-training method, device and electronic equipment proposed in this application are used to solve related technologies. Since the pre-training model pays more attention to the versatility of downstream tasks, it lacks the ability to model tasks in specific directions, resulting in The sentiment analysis model generated by the transfer learning of the pre-training model has a poor effect on text sentiment analysis

Images