Design of Intelligent Question-Answering System Based on Large Language Model

The progress of deep learning and natural language processing technology has promoted the question answering system into a new era, which can understand complex questions more accurately, support multi-source knowledge expression and multi-round interaction, and realize intelligence. As a pre-trained language generation model, LLM can better handle the task of understanding and generating natural language, so as to effectively reduce the development cost of intelligent question answering system and improve its accuracy. In this paper, the BERT model has the characteristics of multi-layer bidirectional Transformer coding structure, and the LLM for intelligent problems is constructed, and the weight clipping optimization method is studied. By deleting neurons with zero weight value in the model, the number of parameters in the model is reduced. Finally, the experimental environment is set up, simulation experiments are carried out, and the original model is compared with the optimized model, which verifies the scientific nature of the optimized model. The LLM studied in this paper improves the accuracy of intelligent problem and provides a new idea for constructing intelligent problem system.