Design And Implementation Of An Automaticquestion Answering System For Open Domain

With the rapid growth of data volume on the Internet,the traditional keyword-based information retrieval method has difficulty in satisfying users because of its low accuracy,large amount of information redundancy,and the need for users to filter a large number of documents.The need in information retrieval.The emergence of the automatic question answering system enables the computer to directly query the natural language questions input by the user,and to perform natural language understanding and analysis on the sentences input by the user,thereby returning simple and accurate answers,so that the user's query efficiency is obtained.Greatly improved.Therefore,the automatic question answering system has become a common research direction and core task in the field of natural language processing and information retrieval.This paper mainly combines the existing research and technical achievements,and studies and implements the system construction,question analysis and candidate answer scoring model of the automatic question answering system.The research results are as follows:1.Use the Java language and maven environment to build a question and answer system.The first is the construction of the database.The database creates two tables,one for the evidence table to store the evidence fragments,one for the problem table for storing the questions to facilitate the direct extraction of the answers when the duplicates occur;the data in the database is mainly in the user's use of the system The <question,answer sentence> pairs that are matched at the time are stored in the database in real time and updated in real time.2.Determine the category to which the problem belongs.The system pre-defines the matching pattern of several types of problems,and matches the pattern with the question,and the currently identifiable problem class name,place name,institution name,and so on.3.Establish a candidate answer scoring model to score the quality of the candidate answers.Among the candidate answer scoring models,there are seven basic evaluation methods,namely,word frequency based evaluation model,vocabulary based evaluation model,shortest vocal based evaluation model,text alignment based evaluation model,and relaxed text alignment based evaluation model.The evaluation model based on the back-to-back text alignment and the evaluation model based on the hot words;the comprehensive evaluation method is a weighted combination method based on the above seven evaluation methods.In order to improve the method of the above method,the method of step-by-step extraction of the answer is to extract the sentence containing the answer from the answer segment,and then extract the final answer from the extracted answer sentence.In the process of extracting the answer sentence,the combination of Bi-LSTM and Max pooling is used to construct the answer extraction model.The experimental results show that the model has achieved considerable results in the extraction of the answer sentence,and the MRR index has reached 0.75.