Automatic Question Answering Method And Application Research Based On Knowledge Graph

With the development of robot technology,robots become more and more intelligent and multifunctional.However,they still lack the ability to understand and respond to natural language,which limits their usefulness in various applications.Therefore,automatic question answering technology has become an important research direction in robot engineering.In the face of complex natural languages and growing Internet information,the current mainstream solutions are based on relational databases,templates and knowledge graphs.However,the performance of the relational database-based method is poor when dealing with multi-table joint queries,and the template-based method requires a large amount of manual annotation to construct rules.Compared to the previous two technologies,knowledge graph-based question answering has good scalability and does not require the manual setting of fixed rules,making it more flexible.This method aims to help users answer natural language questions by storing facts in a knowledge graph without the need to understand its data structure.This is a challenging task because,on the one hand,machines find it difficult to understand the semantics of questions correctly,and on the other hand,current methods cannot effectively utilize the content in the knowledge graph.Specifically,due to the existence of abbreviations,ambiguities,complex grammar,and other factors in questions,especially when the input content contains images,the model’s attention often cannot focus on the correct content,leading to errors in the reasoning chain during retrieval and resulting in difficulty applying the question answering system.In this article,an automatic question answering robot prototype system is designed using syntactic analysis and multi-modal attention to address the above issues.The effectiveness of the proposed method is demonstrated through experiments.The main work and innovation of this paper are as follows:(1)In order to address the problem of word abbreviations and ambiguities in questions,the knowledge graph is embedded in the same space as the word features.The learned vector representations can naturally distinguish the problems caused by abbreviations and ambiguities.In the case of complex grammar,syntactic analysis and graph convolutional neural networks are used to model long-range dependencies between words,better helping machines understand the content of questions.The proposed method has achieved good performance on the Simple Question dataset.(2)When input contains images,a unique multi-modal receptive field is created in this article.Guided by the question,we extend the algorithm’s attention to the image and knowledge graph modals to obtain relationship vectors,in order to make the model more comprehensive.Establish a scientific inference chain based on the entities in the question and image,as well as the relationship vector containing three modalities,to obtain the final answer.The proposed method has achieved good performance on the FVQA and ZS-F-VQA dataset.(3)This article applies the above methods to practical question answering scenarios,where users input images and questions.The system can accurately understand the user’s intentions,reduce the interference of redundant and irrelevant results,provide corresponding answers,improve user query efficiency and accuracy,and greatly enhance the user’s search experience.This algorithm can obtain multi-source knowledge from the knowledge graph and present it to users in a concise and understandable form,which is conducive to the dissemination and popularization of knowledge,further proving the practicality of this method.