Semantic Search by Latent Ontological Features

TL;DR

This paper introduces ontology-based extensions to the Vector Space Model that incorporate latent ontological features of named entities, improving search quality over traditional keyword-based methods.

Contribution

It presents novel models that combine ontological features with keywords for enhanced text retrieval and document representation.

Findings

Improved search quality on benchmark datasets

Advantages in document and query representation

Effective integration of ontological features

Abstract

Both named entities and keywords are important in defining the content of a text in which they occur. In particular, people often use named entities in information search. However, named entities have ontological features, namely, their aliases, classes, and identifiers, which are hidden from their textual appearance. We propose ontology-based extensions of the traditional Vector Space Model that explore different combinations of those latent ontological features with keywords for text retrieval. Our experiments on benchmark datasets show better search quality of the proposed models as compared to the purely keyword-based model, and their advantages for both text retrieval and representation of documents and queries.