Knowledge Map Construction and Question Answering System Design Based on NLP and Neural Network Algorithm

With the explosive growth of information and the continuous evolution of users’ needs, it has become an important research field to build an intelligent Question answering system (QAS) to extract knowledge from large-scale text data. This study combines natural language processing (NLP) technology and neural network model to improve the accuracy and usability of QAS. Firstly, through NLP technologies such as entity recognition, relationship extraction and semantic analysis, a knowledge map is constructed to transform text information into a machine-understandable form. This provides a powerful knowledge base for QAS. Secondly, according to the requirements of QAS, this experiment uses pipeline extraction as the method of entity relationship extraction. In order to improve the accuracy of the relationship classification stage, this experiment added the artificial feature fusion stage after the entity extraction stage to make full use of the entity information. BERT pre-training model is used as the word vector part of this experiment. The research results show that the combination of Bidirectional LSTM (BiLSTM) plus Convolutional Neural Network (CNN) and Self-attention Network based on BERT has achieved the best results, with the accuracy of 82.9% and the F1 value of 74.6%. After using BERT pre-training model, the prediction effect has been significantly improved. Through the system test, the function and performance of the QAS designed and implemented in this paper are in line with the expected results.