Improving approximation of domain-focused, corpus-based, lexical semantic relatedness

Abstract

Semantic relatedness is a measure that quantifies the strength of a semantic link between two concepts. Often, it can be efficiently approximated with methods that operate on words, which represent these concepts. Approximating semantic relatedness between texts and concepts represented by these texts is an important part of many text and knowledge processing tasks of crucial importance in many domain-specific scenarios. The problem of most state-of-the-art methods for calculating domain-specific semantic relatedness is their dependence on highly specialized, structured knowledge resources, which makes these methods poorly adaptable for many usage scenarios. On the other hand, the domain knowledge in the fields such as Life Sciences has become more and more accessible, but mostly in its unstructured form - as texts in large document collections, which makes its use more challenging for automated processing.

In this dissertation, three new corpus-based methods for approximating domain-specific textual semantic relatedness are presented and evaluated with a set of standard benchmarks focused on the field of biomedicine. Nonetheless, the proposed measures are general enough to be adapted to other domain-focused scenarios. The evaluation involves comparisons with other relevant state-of-the-art measures for calculating semantic relatedness and the results suggest that the methods presented here perform comparably or better than other approaches.

Additionally, the dissertation also presents an experiment, in which one of the proposed methods is applied within an ontology matching system, DisMatch. The performance of the system was evaluated externally on a biomedically themed ‘Phenotype’ track of the Ontology Alignment Evaluation Initiative 2016 campaign. The results of the track indicate, that the use distributional semantic relatedness for ontology matching is promising, as the system presented in this thesis did stand out in detecting correct mappings that were not detected by any other systems participating in the track.

The work presented in the dissertation indicates an improvement achieved w.r.t. the stat-of-the-art through the domain adapted use of the distributional principle (i.e. the presented methods are corpus-based and do not require additional resources). The ontology matching experiment showcases practical implications of the presented theoretical body of work.