Popularity-Driven Ontology Ranking Using Qualitative Features

Abstract

Efficient ontology reuse is a key factor in the Semantic Web to enable and enhance the interoperability of computing systems. One important aspect of ontology reuse is concerned with ranking most relevant ontologies based on a keyword query. Apart from the semantic match of query and ontology, the state-of-the-art often relies on ontologies' occurrences in the Linked Open Data (LOD) cloud to determine relevance. We observe that ontologies of some application domains, in particular those related to Web of Things (WoT), often do not appear in the underlying LOD datasets used to define ontologies' popularity, resulting in ineffective ranking scores. This motivated us to investigate -based on the problematic WoT case -whether the scope of ranking models can be extended by relying on qualitative attributes instead of an explicit popularity feature. We propose a novel approach to ontology ranking by (i) selecting a range of relevant qualitative features, (ii) proposing a popularity measure for ontologies based on scholarly data, (iii) training a ranking model that uses ontologies' popularity as prediction target for the relevance degree, and (iv) confirming its validity by testing it on independent datasets derived from the state-of-the-art. We find that qualitative features help to improve the prediction of the relevance degree in terms of popularity. We further discuss the influence of these features on the ranking model.

LOVBench: Ontology Ranking Benchmark

“…In TermPicker [38], a dataset is automatically derived from the Linked Open Data (LOD) cloud, however, relevance judgments only indicate whether a certain triple pattern appears in the LOD cloud or not (binary relevance) and it does not contain any judgments for keyword-based queries. Kolbe et al [21] propose a dataset derived from scholarly data, which, however, only considers ontology and no term ranking, is also limited by its size (1,028) and focuses on IoT ontologies.…”

“…Compared to obtaining explicit judgments from human experts, which may contain bias and become outdated, collecting implicit user feedback through logs is cost-efficient and further captures actual, potentially timevarying user preferences [42]. This helps to overcome the problem that the amount of training data in LTR settings is often low [21,40]. However, clicks cannot be used as direct relevance judgments due to user's inherent biases (e.g., position bias), click incompleteness (missing feedback for relevant terms), and noise (a click does not necessarily imply relevance of a term) [ Experimental results Figure 1: Overview of the LOVBench approach.…”

“…Such features are often considered in ontology ranking by formulating scores involving user feedback such as ratings and clicks [26], ontology and term usage in LOD [38], etc. While these features can be very useful for ranking, applying these features to different ontology collections can be problematic because the respective metadata is often not available or difficult to obtain for other ontology collections [21]. Since ranking models evaluated by LOVBench should be applicable to any ontology collection, we choose to exclude them.…”

“…RDF graph analysis (Feature [17][18][19][20][21][22][23][24][25][26][27][28][29][30][31][32][33]. Features based on the RDF graph O extract statistics about terms in an ontology which, e.g., can give an indication about the importance of terms in the repository and to which detail a term is defined.…”

“…However, using metadata-based features in the ranking model in general can be problematic, depending on the way the metadata is collected. Instead, sufficient and well-designed ontology and term features directly derived from the ontology collection might be able to substitute such features [21].…”