Research On Key Algorithms Of Deep Question Answering System Based On Knowledge Graph Of Exploration And Development

In the work of exploration and development,it is often necessary to retrieve useful information from a large number of unstructured texts.But it is difficult for a full-text search system to fully satisfy the search requirements by using a string matching method.However,based on the knowledge graph,the question answering system can directly output accurate answers according to the user's input questions.The system can meet the efficient retrieval requirements in the field of exploration and development.This thesis studies and designs a deep question answering system based on knowledge graph of exploration and development.According to the type of answer,the deep question answering system is divided into simple question answering and complex question answering,and their respective key algorithms are explored.The problem of simple question answering refers to that the answer corresponding to a question is a factual phrase.In the simple question answering,the important step is the semantic matching of the question and the relation.Due to the small scale of labeled data in professional domain,traditional supervised learning methods are difficult to obtain the high accuracy of semantic matching.To solve this problem,this thesis proposes a transfer learning model based on recurrent neural network.By pre-learning the semantic distribution of a large number of unlabeled samples in the general domain,it improves the accuracy of semantic matching in the professional domain.Experiments show that the proposed model can significantly improve the accuracy of semantic matching compared with the traditional supervised learning method.The problem of complex question answering refers to that the answer corresponding to a question is a descriptive sentence.In the complex question answering,the core step is to construct a reasonable answer generation model to generate an accurate answer.The traditional model only uses the document and question content as input,and then generates the answer automatically by the end to end network structure,ignoring the high-level semantic feature information in the document.To solve this problem,this thesis proposes a complex answer generation model based on triple information.This model enhances the performance of the document by combining the abstract semantic feature information,such as triple information,with the word embedding layer,and then improves the effect of the answer generation.The experimental results show that the model can improve the accuracy of the answer generation compared with the traditional model without triple information.