Knowledge Graph Fusion Based On Representation Learning:An Algorithm And System Implementation

In recent years,with the development of the semantic web,more and more structural data are published in the form of knowledge graph(KG),and widely used in information retrieval,recommendation system and question answering system.As an important component of semantic data,KG usually contains a large amount of RDF triples with overlapping information,while only a few entities have equivalent links within KGs.To use multiple related KGs simultaneously,entities have to be aligned or merged,and the key technology for this problem is entity fusion.Due to the semantic heterogeneity of data between different KGs,there are many variations and ambiguities in the representation of entities and attributes,which brings a great challenge to the entity fusion technology.The fusion algorithm is mainly based on linguistic features,hierarchy,attribute domain,auxiliary information,machine learning and knowledge representation learning etc.In general,algorithms based on linguistic similarity are difficult to be applied to large-scale datasets,machine learning is relatively flexible,but relies on training data and optimization algorithms,knowledge representation learning can be separated from textual information of entities,and encodes entities according to the structural features of RDF triples.Traditional entity fusion systems provide limited matching algorithms,and lack a user friendly interface,which sets a high threshold for an ordinary user.In this paper,we propose an iterative fusion algorithm via mutual supervision based on knowledge representation learning,which maps the entities and attributes into the same low-dimensional vector space.Compared with traditional methods,it avoids the mix of KGs and realizes the entity fusion between cross-lingual KGs.This paper also implements a network-based entity fusion system with a good interactive interface and detailed operation guide that supports time-efficient online fusion calculations and data transmission.This system offers several fusion algorithms,including linguistic distance metrics,machine learning based on positive samples,and knowledge representation learning.