Research On Knowledge Graph Inference And Merging Technologies For Qusetion Answering

knowledge graph is a graph structure data in which each node usually represents a specific thing,and these nodes are interconnected to form a large graph.After the knowledge graph was proposed,it attracted a lot of attention.Knowledge graph based question answers play an important role in applications such as search and personal assistants.Question answering system based on large knowledge graph is usually split into several steps.This paper mainly studies the techniques used in the question answer system on the knowledge graph on the open field.The main research contents include the following three aspects:1.Research on topic entity words extraction based on feature fusion.The topic entity words refers to the topic thing consulted by the question,and the topic entity words extraction is the process of determining this topic thing.In this paper,the question is represented as vector form,and the language model constructed based on long and short time memory network is used to extract the grammatical features of the candidate entity words.This paper also combines lexical and context information,and ranks all candidate entity words to get the correct topic entity words.2.Research on relation matching method based on attention mechanism.After the topic thing is determined,the knowledge subgraph with the central node can be obtained from the knowledge graph.Relation matching is the process of finding the correct path that matches the question from the candidate relation path on the subgraph.In this paper,we use the deep learning method to learn the semantic vector of the question and the relation path.At the same time,we use the string similarity evaluation methods such as edit distance and the word vector similarity of the word bag to match question and the candidate relation path.This article also uses the attention mechanism to enhance the connection of key words in the question and the relationship path text sequence.Finally,the topic entity word extraction and relationship matching and answer reordering modules are integrated into a complete system to get the answer to the question.3.Research on the method of merging data from multiple large knowledge graphs.The large-scale knowledge map is basically constructed by the algorithm.Due to the imperfection of the prior technology and the incompleteness of the data source information,there are many problems in the content of the knowledge graph.And knowledge graph should be constantly updated and expanded over time.Except the usual knowledge graph construction methods,combining various knowledge graph is also an effective construction method.This paper first introduces the problems that may be encountered when processing knowledge graph data,and then studies the knowledge map merge process such as index construction.