Research And Application Of Academic Knowledge Graph Construction For Technological Literature

With the arrival of Internet big data era,how to quickly retrieve papers from massive technological literature,accurately extract domain knowledge and help scholars discover new tasks and methods has become a major challenge for academic community.Academic knowledge graph integrates the heterogeneous information of academic community into a structured knowledge network,and links closely related scientific entities in real semantics,which can help to retrieve papers.A literature usually propose relevant methods for specific tasks.However,the current academic knowledge graph focuses on information such as authors,domains and keywords,and rarely mines the semantic information such as tasks and methods contained in abstracts.If a knowledge graph can be constructed around tasks and methods,including concepts,authors and institutions,semantic reasoning based on meta-path can help scholars locate research content,explore academic trends and discover social networks.Therefore,the goal of this paper is to build an academic knowledge graph based on tasks and methods,including scholars,institutions and keywords.This paper is an important part of "Research,development and application of fast knowledge sharing system in the era of big data and mobile Internet",which is Jilin province's key scientific and technological research and development project.The project aims to develop a mobile learning software for academic knowledge's sharing.Users can read academic information such as venues,literature and author,track technological news and hotspots,and add friends to communicate or comment online.Based on this project,the main work of this paper is as follows:(1)A semi-automatic algorithm of ontology construction based on Wiki data is proposed.It extracts concepts from encyclopedia with high-quality source Wikipedia,and defines nodes' hypernym and hyponym relation.Under the guidance of experts,ontology checking is carried out to prune,clean and complete the raw ontology,so as to build a high-quality ontology to link the entity layer of knowledge graph.(2)An entity and relation jointly extraction algorithm based on Double QA-IE is proposed.It transfers spans extraction task from question answering to knowledge graph construction,extracts entities and relations from abstracts,and links to ontology.Given questions,abstracts and ontologies,candidate ontologies are jointy embedded with text as anchor nodes,to calculate the compatibility of matrix to obtain the attention of words,and to design a bidirectional deep attention network for text representation.In first turn,it jointly predicate start and end positions,type and ontology of entities;in second turn,it predicate relation,tail entities and ontology labels after answers filling.(3)An academic knowledge graph is constructed based on the above algorithm and applied to the practical project.This knowledge graph is a three-layer structure centered on tasks and methods.The entity layer is the core of the graph,and tasks and methods extracted by Double QA-IE algorithm are the central nodes.There are three kinds of relations among the nodes: use,compare and part-of.The academic ontology layer is the basic part of the graph,centered on the scientific concepts extracted by the ontology construction algorithm and used for linking entity layer to add hierarchy relation and resolve ambiguity.The text layer is the extension of the graph,which centers on the literature integrated by the project,containing attribute such as title,keywords and time,and heterogeneous nodes such as venues,authors and users.It is used to locate the text of task and method nodes in the entity layer.The academic knowledge graph constructed in this paper has been applied to the software "Academic Headlines App" supported by the project.