LEAPME: Learning-based Property Matching with Embeddings

作者：

Highlights：

• Integrating data properties from heterogeneous sources going beyond name similarity.

• Instance features give information about format.

• Word-embedding features give information about semantics.

• Supervised training allows the exploitation of the features’ potential.

• There is potential for training a universal, context-independent model.

摘要

•Integrating data properties from heterogeneous sources going beyond name similarity.•Instance features give information about format.•Word-embedding features give information about semantics.•Supervised training allows the exploitation of the features’ potential.•There is potential for training a universal, context-independent model.

论文关键词：Data integration,Machine learning,Knowledge graphs

论文评审过程：Received 9 October 2020, Revised 20 July 2021, Accepted 20 October 2021, Available online 11 November 2021, Version of Record 22 November 2021.

论文官网地址：https://doi.org/10.1016/j.datak.2021.101943