Design And Implementation Of Question Answering System Based On Chinese Knowledge Graph

With the popularity of the Internet technology,the information on the Internet has shown exponential growth in recent years.These colorful information enhances the possibility of people acquiring most of the knowledge in the Internet world,but it also makes the efficiency that people obtain effective information get lower and lower.So the people's voice for getting information quickly is getting higher and higher.The Question Answering system came into being.It is different from the traditional search engine that returns a series of documents or web pages.It can understand the user's search intent and give a concise answer in a very short time.Thus,the Question Answering system becomes one of the most popular artificial intelligence research.The techniques of entity and relation extraction,candidate entity ordering and knowledge representation learning are deeply studied based on the analysis of the related techniques of the knowledge graph question and answer system.Then the innovative models and algorithms are given to improving on traditional classical models and algorithms,which are used to construct the question and answer system based on Chinese knowledge graph.The specific works of the thesis are as follows:(1)The information extraction task is divided into two sub-tasks in traditional serial-methods named entity extraction and relationship extraction.This method has the disadvantages of error accumulation and low information utilization.Therefore,the BA-IE model is proposed by using Bi-GRU to enhance the training speed of the model and adding the attention mechanism to enhance the model's ability to understand the semantics of the questions.Finally,the joint extraction model of entities and relationships is realized.Meanwhile,the better performance of this model in the information extraction task is proved by comparing and verifing under the public dataset.(2)In practical applications,the lists of candidate entities returned by entity-linkers are unordered and non-semantic,which greatly increases the time cost of knowledge reasoning.To sort the lists of entities,the MS-Rank algorithm based on a multi-grade scoring mechanism is proposed,which is composed of popularity,semantic similarity and character similarity.And the algorithm reduces the redundant calculation in the process of knowledge inference.(3)In order to solve the highly similar problem of the knowledge representation of some entities in the traditional TransE model,the fact descriptions of entities are added into the TransE model and an improved FD-TransE model is proposed,expanding the distance between the representations of entities.The result of verification on public dataset proves that the model can effectively solve the above problem.(4)Combining the models and algorithms in Chapters 3 and 4,a Question Answer system based on Chinese knowledge graph named CN-DBpedia is designed and implemented,which finally tests and analyses in the real data flow environment.