Research On The Key Techniques Of Automatic Question Answering For Free Text In Specific Fields

Machine reading comprehension is a complex task in the automatic question answering task.Machine reading comprehension is a technique in which a computer reads a text,understands the content of the text,and finds or deduces the answer to a question.Reading comprehension is also one of the test subjects of human languages.In the process of advanced reading,human beings not only need to understand the original text,but also need further analysis and synthesis.Thanks to the rapid development of deep learning technology,various large data sets based on machine reading comprehension tasks are constantly launched,which makes new progress in machine reading comprehension tasks.Despite the rapid development of machine reading comprehension,the mainstream machine reading comprehension models have been able to surpass the human level in answering the extraction machine reading comprehension task,but these mainstream machine reading comprehension models still remain at the level of semantic matching,which is far from the real "understanding".For more complex reading comprehension tasks,the results were not as good.In this paper,according to the problems faced by the current machine reading comprehension task and the characteristics of the data set of the reading comprehension task in the specific domain used in this paper,a targeted machine reading comprehension answer extraction method is designed.First of all,this paper classifies a complete set of machine reading comprehension data(including various types of extracted machine reading comprehension data,including single-answer questions,multiple-answer questions and reasoning questions according to the types of answers).After the data set is classified,according to the data characteristics of multi-answer questions,some structural characteristics of multianswer questions are used to decompose the data into single answers,and then the single answers are extracted by the benchmark machine reading comprehension model.The final answer for reasoning problem solving,according to answer more questions data format will be reasoning problem decomposition,will be a complete reasoning problem is decomposed into two sub-problems,among the first to use the longest common subsequence obtain annotation data,after using Bi-LSTM training a child problem resolution model,aiming at the problem of separation of the first part of the problem in order to solve the bridge entity,the second part of the problem by getting bridge entity to solve the final answer.