Mongolian multi-modal sentiment analysis method based on T-M BERT pre-training model

Abstract

A Mongolian multi-modal sentiment analysis method based on a T-M BERT pre-training model comprises the steps that neural machine translation and manual correction processing are conducted on Chinese sentiment corpora containing texts, emoticons and GIF short videos, and Mongolian sentiment corpora are obtained; extracting emotional features of Mongolian texts and emoticons by using T-M BERT; the method comprises the following steps: for a Mongolian GIF short video, extracting emotional features by using a G-Transform; and introducing an attention mechanism to dynamically adjust the weight information of the text, the emoticon and the GIF short video to obtain a final emotion feature. And classifying the emotion features by adopting a Softmax function to obtain a final Mongolian multi-modal emotion analysis model, and obtaining an emotion classification result. Finally, the analysis result of the model and the analysis result of a single network are compared and evaluated according to the accuracy rate, the precision rate, the recall rate and the F1 value of each emotion category, and the purpose of improving analysis and public opinion prediction performance is achieved.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence, and in particular relates to a Mongolian multimodal sentiment analysis method based on a T-M BERT (Traditional Mongolian Bidirectional Encoder Representation from Transformers, T-M BERT) pre-training model. Background technique [0002] With the rapid development of Internet technology, people's participation in the Internet is getting higher and higher. The data generated by network users is all-encompassing, including text, emoticons, short videos and other data forms. In the information age, rich channels of information dissemination provide convenience for netizens to publish their views and opinions with personal emotions, making network public opinion have a good interaction. And some negative emotions will have a negative impact and even trigger group panic. [0003] With the rise of artificial intelligence, deep learning methods have received widespread attention. Becaus...

AI Technical Summary

Problems solved by technology

[0004] In order to overcome the above-mentioned shortcoming of the prior art, the object of the present invention is to provide a kind of Mongolian multi-modal emotion analysis method based on T-MBERT pre-training model, has the following three characteristics: first, for Mongolian text and emoji features, the regularized Mongolian word segmentation technology will be used to segment the data, and the trained Vocab dictionary will be corrected by using the Mongolian sentiment dictionary and emoticon dictionary, so as to better alleviate the non-login due to the complexity of Mongolian grammar word problem; secondly, by using T-MBERT and G-Transformer networks to learn the emotional features of three modalities of Mongolian text, emoji and GIF short video respectively

Images