LEAPME: Learning-based Property Matching with Embeddings

Abstract

Data integration tasks such as the creation and extension of knowledge graphs involve the
\nfusion of heterogeneous entities from many sources. Matching and fusion of such entities require
\nto also match and combine their properties (attributes). However, previous schema matching
\napproaches mostly focus on two sources only and often rely on simple similarity measurements.
\nThey thus face problems in challenging use cases such as the integration of heterogeneous
\nproduct entities from many sources.
\nWe therefore present a new machine learning-based property matching approach called
\nLEAPME (LEArning-based Property Matching with Embeddings) that utilizes numerous features
\nof both property names and instance values. The approach heavily makes use of word
\nembeddings to better utilize the domain-specific semantics of both property names and instance
\nvalues. The use of supervised machine learning helps exploit the predictive power of word
\nembeddings.
\nOur comparative evaluation against five baselines for several multi-source datasets with
\nreal-world data shows the high effectiveness of LEAPME. We also show that our approach is
\neven effective when training data from another domain (transfer learning) is used.

Topics & Concepts