Using Contrastive Learning to Optimize the Intention Recognition Model of Sentence Vector Plus CNN Network
Junming Liu, Haiyan Zeng · 2024
With the continuous development of question answering systems, enabling computers to accurately understand the intentions of user questions is of great significance for the entire field of question answering systems. The main goal of intention recognition is to determine the user's intention in the human-machine dialogue process, improve the accuracy and naturalness of the question answering system, and generally achieve it through classification models. This article proposes a new neural network model based on the widely used classification models currently. The sentence vectors generated by the pretrained model are optimized using comparative learning to form a new pretrained model, followed by CNN network feature extraction to improve the recognition ability of the entire model. The proposed neural network model has a certain performance improvement compared to other models, achieving good performance.