Research And Application Of Multi-perspective Academic Knowledge Graph Construction Method For Semantic Reasoning

Knowledge is the development power of human innovation.People discover,research and innovate knowledge in practice,and humans are constantly transforming society through knowledge.Nowadays,information in the online world is complicated,and many redundant and repetitive information interferes with people's sight.Therefore,it is a very important task for academics to quickly and accurately locate their knowledge of interest.Search and recommendation can help people achieve this goal.However,in the past,search and recommendation are generally only from a statistical point of view,and the items that are often clicked have a higher weight,rather than from the perspective of human understanding,to really get semantically relevant knowledge.The emergence of the knowledge graph has solved this problem.The knowledge graph integrates the entities on the network into a semantic network.Through path inference on the knowledge graph,the system can show the entities that are closely related to the real semantics.In the field of academic research,academic papers are the central source for people to acquire knowledge,and academic papers aim to solve academic problems.In order to solve various problems,innovative methods are also the main contribution of the paper.It can be seen that the problems and methods are different angles to understand the paper.In the past,few people extracted the problems and methods of the paper as nodes in the knowledge graph,and how to extract these two entities from unstructured text is also a difficult point.At the same time,the general entity extraction algorithm requires a large amount of labeled data,which consumes a lot of manpower and material resources in the case of a large amount of data.Therefore,the goal of this paper is to use a small amount of labeled data to complete the problem and method extraction under a large amount of data,and build an academic knowledge graph that uses the problem and method as nodes to further solve problems such as semantic reasoning.The main work of this article is as follows:(1)A multi-view concept extraction algorithm based on graph neural network is proposed.Many scholars have successfully constructed an academic knowledge graph,but there are fewknowledge graphs that extract problems and methods as nodes.However,unlike the existing methods,this paper proposes a hypothesis about domains,problems,and methods,which can accurately distinguish problems and methods,and apply the constraints imposed by the hypothesis in the iterative self-training process to achieve semi-supervised learning.For each sentence in the paper,assign a label to each word in the sentence.The label is the question or method.In the process of learning feature vectors,the context vector representation of the word will be fused with the structure vector representation of the associated keywords in the network.The neural network will be used to learn the fused features,and two different objective functions will be used to make the final prediction;In the process of model self-training,the constraints made by assumptions are added,so that the model can automatically develop towards the more accurate classification results,thereby saving a lot of work on labeling data.(2)An academic paper recommendation algorithm based on capsule network is proposed.The knowledge graph constructed in the previous step is auxiliary information.According to the papers clicked by a user,the semantically closely related papers are obtained as a list of candidate papers in the knowledge graph.The user's click behavior is used to learn the characteristics of the user and the paper.The capsule network is used.The above features are extracted to predict the user's rating of the paper,so as to rank the candidate papers,and select the top ranked papers to recommend to the user.The accuracy of the experimental results is verified on the dataset.