Research On Intelligent Question Answering Based On Knowledge Graph Of Ancient Poetry

With the development of the Internet,digital information has shown an exponential growth trend,which brings challenges for people to obtain the information wanted from massive data quickly and accurately.The traditional way of obtaining information is to use search engine,which returns a large number of related webpages,this results in users needing to spend a lot of effort to filter out the answers they need from the returned webpages.Compared with the traditional search-engine-based information acquisition method,intelligent question answering try to understand user's search intention accurately and return the answer to the user directly,which improves the efficiency of user information acquisition.Furthermore,with the development of knowledge graph,it can be used as a high-quality data source for intelligent question answering,and its rapid development has promoted the application of intelligent question answering in multiple fields.At present,the learning and application of ancient poetry knowledge is very important in the field of education.However,the system of ancient poetry knowledge is complex.The method for users to obtain ancient poetry information is mainly through a search engine,this information acquisition method is not efficient enough for those users who want to get ancient poetry information according to their own need.Therefore,constructs a knowledge graph of ancient poetry,and implements an intelligent question answering about knowledge of ancient poetry based on the knowledge graph.The research mainly includes the following four sides:First,constructing a knowledge graph of ancient poetry.Based on the data in the relational database,supplemented by the Internet data,fuses the data from different sources to construct an ancient poetry knowledge graph which can be used for intelligent question answering.Second,studying question classification algorithm based on BERT.Most of the commonly used classification algorithms use Word2vec to obtain the word vector representation of the text.This word vector representation method has certain limitations,the word vector after training is universal and won't change.But the semantics of the same word in different contexts is different.Therefore,use BERT to obtain word vectors which contains contextual semantic information,and implement the classification of user's question through BERT.Experimental results show that the effect of question classification using BERT has been improved.Third,studying the entity recognition algorithm(BiLSTM-CRF)based on the bidirectional long-short memory network(BiLSTM)with conditional random field(CRF)added.In the actual context,the context information of a single word will affect the semantics of the word,and traditional neural networks cannot capture long-distance context information.To solve this problem,use BiLSTM-CRF to identify the entities of question.Firstly,the bidirectional long-short memory network(BiLSTM)can solve the long-distance text dependence problem on a certain scale.Secondly,combined with the conditional random field(CRF)to obtain the entity label's existing dependency relationship and ultimately improves the entity recognition effect.In addition,based on BiLSTM-CRF,verify the influence of different word vectors to the algorithm.Experiments prove that BiLSTM-CRF using BERT to pre-training word vectors can obtain the better results.Fourth,implementing an intelligent question answering system based on the knowledge graph of ancient poetry.In the actual working process,the intelligent question and answering system is in a good operating condition and can answer the user's questions accurately,satisfies the user's needs for obtaining ancient poetry information.