Aiming at non-topic entity and multi-topic entities,fuzzy topic complex questions,this paper proposes an im-proved scheme based on the staged query graph generation.For non topic entity complex questions,this paper proposes a query graph construction scheme with the answer node as the starting node.For multi topic entities complex questions,this paper proposes a topic entity selection strategy based on rules and Transformer encoder,which improves the accuracy of the system.The compara-tive experiments on four data sets show that this method performs better than the recent better Multi-hopCQA on complex questions.
维普
万方数据