Authors
Nitisha Jain, Ralf Krestel
Abstract
Knowledge graphs provide a structured representation of data in the form of relations between different entities. The semantics of relations between words and entities are often ambiguous, where it is common to find polysemous relations that represent multiple semantics based on the context. This ambiguity in relation semantics also proliferates knowledge graph triples. While the guidance from custom-designed ontologies addresses this issue to some extent, our analysis shows that the heterogeneity and complexity of real-world data still results in substantial relation polysemy within popular KG datasets. Yet, this issue has been largely overlooked in the past. The correct semantic interpretation of knowledge graph relations is necessary for downstream applications such as entity classification, question answering and knowledge graph completion.
We present the problem of fine-grained relation discovery and a data-driven method towards this task that leverages the latent vector representations of the knowledge graph entities and relations available from relational learning models. Our method shows that by way of performing clustering over these embeddings, it is possible to not only identify the polysemous relations in knowledge graphs, but to even discover the different semantics associated with such relations for deriving fine-grained relations. Extensive empirical evaluation demonstrates that fine-grained relations discovered by the proposed approach lead to substantial improvement in the relation semantics in the Yago and NELL datasets, as compared to several related baseline approaches. In addition, we present a qualitative analysis as well as the insights from manual evaluation that convey that fine-grained relation discovery is an important yet complex task, especially in the presence of hierarchical ontologies and sparsity of data.
