Path-enhanced Multi-hop Question Answering Method Based on Knowledge Graph Embedding
A path-enhanced multi-hop question answering method based on knowledge graph embedding is proposed. A "comparation-aggregation" framework is constructed,which can enhance the question-to-answer path to solve the problem of poor question answering effect caused by not making full use of the path and natural language question answering order in previous models. Different comparison functions are designed in this paper,and SUBMULT+NN is chosen as the comparison function. Experiments were carried out on MateQA and WebQuestionsSP,two benchmark question-and-answer datasets in open fields. Compared with previous multi-jump question-and-answer methods,the results are improved on the complete knowledge graph and excellent on the incomplete knowledge graph.
万方数据
维普
