[Paper Review] Enriching Linked Datasets with New Object Properties

Srinivasagan N. Subhashree, P. Sreenivasa Kumar|arXiv (Cornell University)|Jun 24, 2016

Semantic Web and Ontologies30 references3 citations

TL;DR

DART is an unsupervised framework that discovers new object properties between two given LOD classes by leveraging contextual similarity and paraphrase detection on web text patterns. It outperforms existing systems like newOntExt in both correctness and quantity of discovered relations, while automatically grounding results to eliminate irrelevant relations and generate candidate property axioms for T-Box enrichment.

ABSTRACT

Although several RDF knowledge bases are available through the LOD initiative, the ontology schema of such linked datasets is not very rich. In particular, they lack object properties. The problem of finding new object properties (and their instances) between any two given classes has not been investigated in detail in the context of Linked Data. In this paper, we present DART (Detecting Arbitrary Relations for enriching T-Boxes of Linked Data) - an unsupervised solution to enrich the LOD cloud with new object properties between two given classes. DART exploits contextual similarity to identify text patterns from the web corpus that can potentially represent relations between individuals. These text patterns are then clustered by means of paraphrase detection to capture the object properties between the two given LOD classes. DART also performs fully automated mapping of the discovered relations to the properties in the linked dataset. This serves many purposes such as identification of completely new relations, elimination of irrelevant relations, and generation of prospective property axioms. We have empirically evaluated our approach on several pairs of classes and found that the system can indeed be used for enriching the linked datasets with new object properties and their instances. We compared DART with newOntExt system which is an offshoot of the NELL (Never-Ending Language Learning) effort. Our experiments reveal that DART gives better results than newOntExt with respect to both the correctness, as well as the number of relations.

Motivation & Objective

• To address the lack of object properties in LOD datasets, which limits semantic expressiveness and reasoning capabilities.
• To develop an unsupervised method for discovering new, meaningful object properties between two given LOD classes.
• To automatically ground discovered relations to existing LOD properties, discarding irrelevant or semantically mismatched relations.
• To generate candidate property axioms (e.g., sub-property, equivalent, inverse) for T-Box enrichment.
• To outperform existing systems like newOntExt in both precision and recall of discovered relations.

Proposed method

• DART extracts text patterns from a web corpus that link individuals of two given LOD classes.
• It applies contextual similarity and paraphrase detection to cluster semantically similar patterns into potential object properties.
• The system performs automated grounding by checking domain and range type compatibility with existing LOD properties.
• Grounded relations are validated for semantic relevance, and irrelevant ones (e.g., 'is the father of' between Religions and Countries) are discarded.
• The framework generates candidate property axioms, including sub-property, equivalent, and inverse property relations.
• It supports fully automated discovery and mapping without requiring manual annotation or external knowledge.

Experimental results

Research questions

• RQ1 Can an unsupervised method effectively discover new object properties between two given LOD classes?
• RQ2 How can contextual similarity and paraphrase detection be combined to improve relation clustering accuracy?
• RQ3 To what extent can automated grounding eliminate semantically incorrect or mismatched relations?
• RQ4 How does DART compare to existing systems like newOntExt in terms of correctness and number of discovered relations?
• RQ5 Can DART generate valid candidate property axioms (e.g., sub-property, inverse) for T-Box enrichment?

Key findings

• DART achieved an accuracy of 0.833 for the Rulers-Empires pair, significantly outperforming newOntExt in both correctness and relation count.
• For the Religions-Countries pair, DART identified 25 correct relations with 50% accuracy, demonstrating robustness across diverse class pairs.
• The grounding phase improved performance by discarding semantically mismatched relations such as 'is the father of' between Religions and Countries.
• DART discovered 15 meaningful relations between Writers and Novels, with 52% accuracy, showing effectiveness in domain-specific relation extraction.
• In the Music genres-Music genres pair, DART achieved 69% accuracy and generated 9 grounded relations, including 'is a subgenre of' mapped to the existing property.
• The system outperformed newOntExt in both the number of relations and correctness, particularly in distinguishing semantically distinct relations like 'doesn’t play at' and 'wants to play at'.