Research On Conversational Machine Reading Comprehension Based On Time Sequence Information

Machine reading comprehension is a very valuable research field in natural language processing.Since the birth of the pre training language model,it has become a trend to use the pre training language model to solve the problem of machine reading comprehension and other natural language processing tasks.But for example,Bert model only allows fixed length input,which limits the model's access to and use of temporal information in the text.In addition,the temporal information in historical problems can not be extracted and used well.This paper focuses on temporal information in natural language.The details are as follows:1.Aiming at the limitation of the pre training language model Bert in obtaining the global information of long documents,a machine reading comprehension method based on the sequence information of text paragraphs is proposed.Firstly,the method uses the pre training language model Bert to obtain more accurate word vector feature representation,and the position coding mechanism in Bert is conducive to obtaining fine-grained text temporal position information;Using the timing information extraction module to extract more abundant text timing information to solve the problem of obtaining timing information in text paragraphs;Use the information exchange and transmission module to transmit the timing information between segments,so as to better integrate and utilize the global information of the text.2.This paper proposes a machine reading comprehension model based on the temporal information of historical problems.In addition to paragraphs,questions often contain timing information.In the dialogue reading comprehension task,the model can integrate the information of historical questions into the current questions to be answered.The existing work integrates the historical problem information into the text through the splicing method,which can not effectively exchange information.The model based on the temporal information of historical problems first obtains the word vector representation of historical problems,current problems and articles through Bert model,and then uses Bi-LSTM?Attention network carries out more in-depth interaction with historical issues,current round issues and texts.Due to BiLSTM?Attention network combines the advantages of long-term and short-term memory network and attention mechanism.It can filter out irrelevant information and integrate the useful information in historical problems into current problems.It solves the problem of filtering and using temporal information in historical problems.The experimental results of the reading comprehension method based on text paragraph timing information on coqa dataset show that the F1 value of the model is improved by 1.5%compared with the benchmark comparison model.The experiment of machine reading comprehension model based on the temporal information of historical problems was carried out on two dialog data sets(Co QA,Qu AC).The experimental results show that the model is effective in obtaining the temporal feature information of historical problems.