Research on emotion recognition method based on multi-task learning model

Text emotion recognition is an important task in natural language processing, and plays an important role in government public opinion recognition, commodity evaluation, and social media opinion mining. Emotional information is of great significance. Traditional emotion recognition relies on emotion dictionaries and lacks certain generalization ability. The deep learning-based method requires a large amount of labeled data. a small amount of data, which is easy to cause overfitting. This paper proposes a multi-task learning model based on Chinese sentiment dictionary-assisted training. The model integrates chinese character information and word information, which can improve the generalization ability and avoid overfitting. The feasibility of the model is verified on the weibo_100k data. The results show that compared with the pre-training model, the multi-task learning model in this paper fully taps the coding ability of the transformer encoder, and the multi-task learning model is compared with the single-task model. The performance is improved and the risk of overfitting is significantly reduced. The integrated word information and character information enhance the semantic ability of text information and improve the generalization ability of the model under the multi-task learning framework.