Research And Implementation Of Intelligent Question Answering Technology For Complex Questions

With the development of information technology,knowledge base has become an important form of knowledge storage,but ordinary users can not directly use natural language questions to query answers in the structured knowledge base.As an interface between users and the knowledge base,the intelligent question answering system can understand the natural language questions raised by users and accurately query the answers in the knowledge base.It is one of the research hotspots in the field of natural language processing.When the system needs to answer complex questions raised by users,the diversity of entities,relationships and structures in complex questions poses a challenge to question answering system’s question understanding and knowledge reasoning capabilities.This thesis focuses on the complex questions answering over knowledge base,which divides the task into two parts: question parsing and formal query generation.It proposes a multi-task learning question parsing model and a case-based reasoning-based formal query generation model,and uses these two models to construct an intelligent question answering system.The specific contributions of this thesis are as follows:(1)A complex question parsing model based on multi-task learning is proposed,which includes entity linking,relation linking,and question structure skeleton generation models.In the entity linking model,entity representation is enhanced by combining adjacent relations,while in the relation linking model,a two-stage detection method based on a dual encoder and cross-encoder architecture is proposed to improve the accuracy of relation linking task from the perspective of coverage and precision.In the question structure skeleton generation model,the correctness of skeleton generation is ensured by adding constraints during the decoding process.Finally,multi-task joint learning is achieved by sharing the encoding layer of the three tasks,and the experimental results demonstrate that the overall performance of the model in question parsing tasks has been improved.(2)A Formal query generation model based on case-based reasoning is proposed.To improve the model’s reasoning ability in answering complex questions,this thesis attempts to apply the case-based reasoning framework to the structured query generation task.The thesis designed similar case retrieval module,input construction module,query generation module,and query correction module,respectively.By introducing multiple granularity corpus information such as question parsing structure and similar question cases into the question encoding process,the pre-trained generation model can more accurately generate structured query statements.Finally,through experiments,it was demonstrated that using a case-based reasoning framework can improve the effectiveness of the structured query generation model.(3)Based on the two models proposed in this thesis,an intelligent question answering system is designed and implemented to provide question answering services for users.The system is mainly based on a browser/server architecture,which can meet practical usage requirements.