A similarity model based on reinforcement local maximum connected same destination structure oriented to disordered fusion of knowledge graphs

摘要

The alignment and fusion of knowledge graphs have been at entity alignment and fusion which match and align knowledge graphs (KGs) by measuring similarity of the entities in KGs. Nevertheless, false fusion of completely different KGs can be easily caused if only considering entity similarity but ignoring entity relationship similarity. This paper focuses on entity relationship similarity model and KGs fusion method to achieve automatic construction of KGs. First, Same Destination Paths is developed based on Maximum Common Subgraph, which is used to build the Local Maximum Connected Same Destination Structure (LMCSDS) model to measure the entity relationship similarity of KGs. Then, a fusion method for similar fragmentation KGs (FKGs) is developed by analyzing the types of FKGs. Third, a Reinforcement Local Maximum Connected Same Destination Structure (RLMCSDS) similarity model is developed to ensure that the similarity between FKGs can still be measured correctly after fusion of FKGs. Meanwhile, the fusion results obtained by the developed RLMCSDS model and fusion method are theoretically proved to be independent with the fusion order of FKGs. Finally, experimental results on several datasets demonstrate the outstanding performance of the RLMCSDS model in comparison with some existing methods. Moreover, a complete KG can be built by the RLMCSDS model and the fusion method.