Research on Intent-Slot Recognition Algorithm Based on Knowledge Graph and User Topic

Abstract

In this article, we propose a joint recognition method of intention and slot. The model can directly obtain the user's intention and slot information through the user's historical interaction information, knowledge graph and current input. In this method, we first use a knowledge reasoning module based on user topic and knowledge graph. The topic model is used to filter the user's historical information, remove meaningless chat information, retain the most important topic information, and further obtain the external information vector from the knowledge graph which is helpful for current intention recognition and slot filling. In addition, we use attention mechanism modules to fuse the historical information, external knowledge vector and sentence features, and suppress the invalid part of external information. Experiments on the data sets show that our method can effectively improve the accuracy of intention recognition and F value of slot filling.