Research And Application Of Question Classification And Answer Selection For Question Answering System

In the context of big data,the rapid development of information technology promotes the explosive growth of Internet information.It has become a challenging business for finding the required knowledge in massive information.The traditional way of information retrieval by search engine is not enough to meet the current needs because it cannot return the only answer.Intelligent question answering system is different from search engines.It can meet people's demand for acquisition of information by giving out accurate answers through natural language input.Question answering system has become an effective means to solve this problem.Intelligent question answering has also become a popular technology in natural language processing.However,there are still many problems in the existing technology for constructing question and answer systems,such as inaccurate question classification and answer selection,etc.In order to solve these problems and build a practical question answering system,the main contribution of the research includes:(1)In terms of question classification in question answering system,an attention-based multifeature fusion method is proposed for question classification in this thesis.In this model,term frequency–inverse document frequency is calculated and used to build feature dictionary.The dictionary is used to vectorize the text and extract the statistical features of the text.Then,we combined it with the features extracted by long short-term memory and convolutional neural networks into multi angle text semantic information.Finally,attention mechanism is used to assign different weights to each feature to stabilize the effect of the model.Experimental results show that our proposed model outperforms traditional machine learning models and mainstream deep learning models on short text classification.(2)In view of the best answer choice is not accurate enough,a multi-angle attention feature matching algorithm is proposed for answer selection in this thesis.We use an encoding layer to enrich text features which includes extracting sequence information through bidirectional long short-term memory and extensive semantic information by different convolution kernels.The attention mechanism is used to extract the interactive information between questions and answers to strengthen the local features of the text in the text similarity matching stage.Then we use the global similarity feature to weight the local features to obtain the final similarity.Experimental results show that our model can get more accurate results compared to traditional algorithms based on feature engineering and other deep learning algorithms.(3)We construct an intelligent question answering system for knowledge acquisition by combining the above algorithms.The system is mainly divided into two modules.Ordinary users can get the knowledge they want through the intelligent question answering module.They can also give feedback on the quality of knowledge or view related questions and answers.The administrator monitors the status of the system and adds content to the knowledge base.This system can help users obtain knowledge efficiently and accurately.