| For academics,thesis is an important way to acquire knowledge.It usually aims at a scientific research problem in a certain field and proposes an innovative method for the problem,which highly condenses the academic achievements of scientific researchers.It can be seen that problems and methods are the core of an academic paper,and more knowledge is derived by constantly solving new problems and creating new methods.However,ordinary academic graphs seldom take problems and methods as entities and actually use them to solve reasoning tasks,and how to extract these two entities and the relationship between them from unstructured text is also a difficult point.Traditional knowledge graph construction methods often separate named entity recognition and relationship extraction and perform them separately in a pipelined method,which leads to ignoring the interdependence between the two subtasks,and it is easy to propagate errors in upstream tasks to downstream tasks.In view of the above problems,the main work of this paper is as follows:(1)A multi-task academic knowledge graph containing questions and methods is designed.This paper takes the problem and method of a paper as the core of the task,extracts the hidden problem(task)and method(method)entities in the title and abstract of the paper,and uses them as two types of implicit nodes in the academic graph,thereby constructing an inclusion A multi-task academic knowledge map of problems and methods,which can solve the reasoning problems that the existing knowledge map cannot handle.(2)A joint extraction algorithm for academic entity relations based on event extraction is proposed.For the first time,the method of event extraction was transferred to the task of constructing academic knowledge graphs.The entity and relationship are modeled as entity and argument roles in the event,and the entity and relationship extraction task is modeled as a sentence structure prediction problem.In this process,entity recognition and relationship extraction tasks are performed alternately,and the mutual dependence can be utilized to the greatest extent,so that academic entities and relationships can be more accurately identified.In addition,in order to eliminate the semantic ambiguity of entities in the graph,this paper constructs an ontology library in the field of artificial intelligence based on Wikipedia.(3)Compared with existing entity recognition and relationship extraction algorithms on multiple real data sets,the experimental results show that the event extraction-based academic entity relationship joint extraction algorithm proposed in this paper performs better than the comparison algorithm.In addition,through sampling and testing,the accuracy of the multi-task academic knowledge graph constructed in this paper,which contains "questions" and "methods",is evaluated in reasoning tasks based on academic entities and relationships.(4)The topic selection of this article is based on the key science and technology project of Jilin Province "Research,Development and Application of Rapid Knowledge Sharing in the Era of Big Data and Mobile Internet".This article applies the above work to the academic headline APP under the project(https: //www.acheadline.com/),mining the problems and methods contained in the paper,and integrating the author,domain,publication place and keywords of each paper in the APP to construct a multi-task academic knowledge graph based on the academic Knowledge graph can solve reasoning problems related to problems and methods. |
