An Interview Text Classification Model Based on Deep Learning

• 1、School of Artificial Intelligence, Beijing University of Posts and Telecommunications, Beijing 100876
• 2、International Department， Beijing Bayi School

Abstract：To address the current issue of insufficient text data labels in deep learning in the medical diagnosis of depression, this paper proposes an emotion classification fusion model based on Bert BiLstm Attention. Firstly, the Bert pre trained language model is used to extract feature vectors, and then a neural network model combining BiLstm bidirectional long short term memory network and Attention attention mechanism is used to train the model on the public dataset of emotional dialogue, obtaining a classification model suitable for depression analysis. On this basis, through deep transfer learning method, the model parameters are transferred to the target domain for training. Through continuous fine-tuning in real medical consultation room dialogue data, the model\'s confusion is reduced and its robustness is improved. Finally, a new emotion classification model suitable for depression medical consultation room dialogue scenarios is obtained. The experimental results show that the F1 value of the target domain dataset has been improved by 4.8% through transfer learning method. The experimental verification shows that the model can achieve good classification performance in interview texts with depressive emotions, and accurately identify depressive emotions in real medical consulting room scenarios, providing important assistance for medical diagnosis.

Keywords： Artificial intelligence Text classification Emotional analysis Depression recognition Transfer learning