Knowledge Extraction And Multi-intention Recognition Based On Knowledge Graph Method Study

At present,the question answering system based on knowledge graph occupies an increasingly important position in the field of artificial intelligence.The tasks involved in question answering system based on knowledge graph have gradually become the focus of research.Therefore,in order to filter redundant information,more intelligently meet user needs and improve work efficiency,this paper studies the tasks involved in the question answering system based on knowledge graph from two aspects: knowledge extraction and multi-intention recognition.First of all,since the input data is mostly unstructured short text data,short text data lacks contextual semantic background knowledge,so it is very challenging to accurately extract the relationship between entities.At the same time,the interaction between entity extraction model and relationship extraction model and the multiple relationships between entities bring great difficulties to knowledge extraction task,which will affect the downstream task execution.Therefore,this paper proposes a joint model of knowledge extraction based on the global attention mechanism.The entity extraction module and the relation extraction module are carried out at the same time,which effectively alleviates the existing problems of the joint model,namely,the error superposition caused by the separation of entity extraction task and relation extraction task.Then,on the basis of extracting entities and relationships,in order to further identify multiple intents contained in the questions raised by users,this paper proposes a multi-intents recognition and slot filling model based on graph attention mechanism and gating mechanism.The multi-intention recognition decoder is used in the intent recognition module to predict multiple intents in sentences.Aiming at the slow decoding speed of the mainstream joint model,the graph attention mechanism is used to construct the graph network structure by combining the predicted multiple intent results with slots,so as to realize parallel decoding of slot filling and improve slot filling speed.At the same time,in order to improve the interaction between the association models,a gating mechanism is used to connect the features between the multi-intention recognition module and slot filling module,so as to improve the recognition accuracy of slot filling in the short text.