基于知识图谱的通用知识问答系统:体系与方法

Knowledge Graph Based Universal Question Answering System:Framework and Methods

近年来,众多研究者在某个特定知识图谱上的问答技术上提出了许多成熟的模型,解决了大量具体的问题。但真实应用场景下,基于知识图谱的问答系统要求能够在多个图谱中自由切换,并且能够回答用户表述各异、带有噪音、带有因疏忽导致的错误等各类问题。满足这种真实场景下的问答系统的相关研究甚少。本文从系统性工程的角度出发,提出了一种不依附于特定知识图谱的通用知识问答体系——渊海知识问答(YHQA)体系,并详细描述了其体系架构,以及各个模块适用的方法。最后,本文结合产业界的应用实践,给出了YHQA体系的实现方法描述、部署方案以及行业应用案例。

In Recent years,many researchers have proposed many mature question answering models applied to specific knowledge graph to solve the concrete problems.However,in the real scenario,a KBQA system is required to be able to switch among multiple graphs freely and answer questions with different expressions,noise,and errors caused by negligence.At present,it is not possible to build a QA system using a single method.From the perspective of systematic engineering,the paper proposes a universal QA framework based on all kinds of knowledge graph rather than a specific one,which is called the Yuanhai Question Answering(YHQA)framework.This paper details the architecture and the appropriate methods of the YHQA framework.