Automatic Question Answering Method Based On Retrieval And Machine Reading Comprehension

Obtaining information is an indispensable part of daily life.However,there are a lot of information on the Internet.It takes time and effort to obtain the information you need.In order to solve the problem,automatic question answering task is proposed.The purpose is to find the answers people need from the Internet.As an important part of natural language processing,automatic question answering has been greatly developed in recent years.Thanks to many question answering data sets,more and more automatic question answering systems achieve good results.Existing methods often assume that the text corresponding to the question already exists,but this is difficult to satisfy in reality,and automatic question answering method proposed in this paper is not subject to this limitation.Due to the complexity of machine reading comprehension tasks,feature-based methods are difficult to construct features manually.The simple model based on neural network is not ideal.The model with better effect has more parameters.The method of mixing features with neural networks requires a good combination of the two aspects.This paper proposes a machine reading comprehension model that combines bidirectional additive attention mechanism and recurrent answer pointer network.The model is moderate,easy to train,and has few parameters.The training only needs 10,000-level corpus to converge,and it does not need to manually construct the required features of the task.Under this background,this article proposes automatic question answering method based on retrieval and machine reading comprehension.The main work and contribu-tions of this article are as follows:1.This paper proposes a machine reading comprehension pipeline combined with bidi-rectional additive attention mechanism and recurrent answer pointer network,which can realize the reading comprehension of input natural language questions and text containing answers to get answers.It mainly includes the word embedding layer,the pre-train presentation layer,context presentation layer,bidirectional attention layer,modeling layer and answer pointer layer.2.Based on the traditional attention mechanism,this paper designs a bidirectional addi-tive attention mechanism suitable for reading comprehension,which can effectively obtain the similarity between the passage and the question,and then obtain the pas-sage to question attention and question to passage attention based on the similarity.,is conducive to the positioning of answers to the reading comprehension model.This article expands the pointer network to a recurrent answer pointer network and applies it to the reading comprehension model,using a soft pointer to output the answer at the start and end of the original text.3.This paper implements a TF-IDF text retrieval based on binary language model and a rerank method of character level n-ary language model matching.Combining text retrieval,rerank and machine reading comprehension model.Combining text re-trieval,rerank and machine reading comprehension models,automatic question an-swering system based on retrieval and machine reading comprehension is realized.The system is able to input questions in natural language and return answers to ques-tions.