Question Classification Method based on Self-Attention Mechanism and BiLSTM-AlexNet Model

With the rapid development of Natural Language Processing and deep learning algorithms, research and applications based on intelligent question answering have gained widespread attention from researchers and the industry. As the primary step in intelligent question answering, the accuracy of question classification is crucial to answer generation. In order to solve the problems of single model and poor classification performance in Chinese question classification, this paper proposes a question classification model based on self-attention mechanism combined with BiLSTM-AlexNet model (named A-Bi-Al) . Firstly, word2vec-based word vector representation is performed, and then the local features of the question text are extracted using convolutional neural network (AlexNet). Secondly,bidirectional long short-term memory network (BiLSTM) is used to extract semantic crrelation information of context to obtain high-level feature expression at the sentence level, then feature weighting is performed through self-attention mechanism to reduce the influence of noise features. Finally, classification is performed through the output layer. In addition, the more smooth and adaptive Mish function is used instead of traditional activation function to enrich local feature representation in AlexNet, BiLSTM, and self-attention layers. Relevant experiments were conducted based on LTP and THUC- News datasets. The experimental results show that the model achieves good performance and can effectively improve the effect of question classification.