Question Answering System In Financial Field Based On Machine Reading Comprehension

In today's era of data and information,the rational use of data and the rapid and accurate acquisition of information from the data are hot issues in today's information technology.Due to the large data size and complex structure,the traditional retrieval question and answer cannot meet the contemporary needs.The characteristics of the question and answer system are simple,fast and accurate to obtain information to comply with the development trend of the times.Its research progress has attracted much attention in the fields of artificial intelligence and natural language processing,and is widely used in e-commerce,medical,education and other applications.The Q & A system in the financial field has gradually developed from bank customer answers to online investment lending Q & A.The research on the Q & A system in the financial field has shown an upward trend.In order to improve the performance of the question answering system in the financial field,improve the adaptability of the machine reading comprehension model to the data in the financial field,make reasonable improvements to the machine reading comprehension model,and combine the machine reading comprehension and question matching technology to build a question answering system in the financial field.The innovations are as follows.(1)On the basis of BM25,the semantic features of the problem are combined,the subject and object semantic component factors are added,and the word order rules between the subject and the object are combined to enhance the semantic connection between sentences and make the similarity calculation between sentences more accurate.An incremental test is added to determine the effect of different data sizes on the performance of the problem matching algorithm.The experimental results show that the improved question matching algorithm has an accuracy rate 17.53% higher than that of BM25 and 18.69% higher than Vmodel.With the increase of data increment,the accuracy rate and F value are more stable,showing that the scale of the data has less influence on the problem matching algorithm in this paper.(2)Based on Match-LSTM,the corresponding data reconstruction strategy is adopted according to the role of different data sets in the model.The articles are reordered according to the relevance to the question or answer,highlighting the characteristics of the articles with greater relevance.The self-attention mechanism of the document deepens the relationship between the problem and the article,highlights the features of the article that are deeply related to the problem,and combines multiple articles to make the self-attention mechanism to highlight the correlation between the semantic features of the article and the article.This paper also analyzes the performance impact of the model under different problem types,and the final verification results reach Rouge-L and Bleu-4 are 44.65 and 38.37,respectively.The model has a relatively high improvement on entity and description problems.(3)Combine the improved question matching algorithm and improved machine reading comprehension model into a question and answer system in the financial field,obtain financial related data from Baidu search according to financial related keywords,and sort the data according to a certain data structure.Under the data,the accuracy of the question answering system in this paper is higher than that of other question answering systems using reading comprehension models.