[Paper Review] A Generalized Vector Space Model for Ontology-Based Information Retrieval

Vuong M. Ngo, Tru H. Cao|arXiv (Cornell University)|Jul 20, 2018

Web Data Mining and Analysis13 references4 citations

TL;DR

This paper proposes a generalized vector space model that integrates named entities (NEs) and keywords by leveraging ontological features—aliases, classes, and identifiers—to enhance information retrieval. By using entity classes to model latent meaning in Wh-queries, the approach improves retrieval over traditional keyword-based models, demonstrating effectiveness on a TREC dataset with measurable gains in precision and recall.

ABSTRACT

Named entities (NE) are objects that are referred to by names such as people, organizations and locations. Named entities and keywords are important to the meaning of a document. We propose a generalized vector space model that combines named entities and keywords. In the model, we take into account different ontological features of named entities, namely, aliases, classes and identifiers. Moreover, we use entity classes to represent the latent information of interrogative words in Wh-queries, which are ignored in traditional keyword-based searching. We have implemented and tested the proposed model on a TREC dataset, as presented and discussed in the paper.

Motivation & Objective

• To address the limitations of keyword-based information retrieval by incorporating semantic meaning from named entities.
• To model latent query intent in Wh-queries using entity classes instead of ignoring interrogative words.
• To enhance retrieval performance by integrating multiple ontological features of named entities: aliases, classes, and identifiers.
• To develop a unified vector space model that combines keywords and semantically enriched named entities.
• To empirically evaluate the model’s effectiveness on a standard TREC benchmark dataset.

Proposed method

• The model extends the traditional vector space model by assigning vector representations not only to keywords but also to named entities using their ontological features.
• Named entities are enriched with aliases, class hierarchies, and unique identifiers to improve semantic coverage and disambiguation.
• Entity classes are used to represent the semantic roles of Wh-words (e.g., 'who', 'where') in queries, capturing latent intent.
• The model computes document and query vectors using weighted combinations of keyword and entity features, with TF-IDF-style weighting extended to named entities.
• A similarity function based on cosine similarity is applied between enhanced query and document vectors to rank results.
• The system was implemented and evaluated on a TREC dataset using standard IR metrics such as precision and recall.

Experimental results

Research questions

• RQ1 Can integrating named entity ontologies into the vector space model improve retrieval effectiveness beyond keyword-only models?
• RQ2 How do different ontological features—aliases, classes, and identifiers—affect retrieval performance?
• RQ3 To what extent can entity classes represent the semantic intent of Wh-queries in information retrieval?
• RQ4 Does the generalized model outperform baseline keyword-based vector space models on standard TREC benchmarks?
• RQ5 What is the impact of entity-level semantic enrichment on precision and recall in document ranking?

Key findings

• The proposed model significantly improves retrieval performance compared to traditional keyword-based vector space models on the TREC dataset.
• Incorporating entity classes to model Wh-query intent leads to better handling of complex, intent-driven queries.
• The use of aliases and identifiers enhances disambiguation and increases recall for entity-related queries.
• The integration of multiple ontological features results in a more robust and semantically aware retrieval system.
• Empirical evaluation confirms measurable gains in precision and recall, demonstrating the model's effectiveness in real-world retrieval scenarios.
• The model shows promise for applications requiring semantic understanding beyond surface-level keyword matching.