[Paper Review] Query Expansion: Term Selection using the EWC Semantic Relatedness Measure

Vitaly Klyuev, Yannis Haralambous|arXiv (Cornell University)|Aug 19, 2011

Information Retrieval and Search Behavior16 references3 citations

TL;DR

This paper proposes a query expansion method using the EWC semantic relatedness measure, which integrates Wikipedia-based Explicit Semantic Analysis, WordNet path similarity, and a mixed collocation index to improve ad-hoc retrieval. Evaluated on NTCIR data using Terrier, the approach significantly enhances retrieval performance, demonstrating the effectiveness of combining multiple semantic signals for term selection in query expansion.

ABSTRACT

This paper investigates the efficiency of the EWC semantic relatedness measure in an ad-hoc retrieval task. This measure combines the Wikipedia-based Explicit Semantic Analysis measure, the WordNet path measure and the mixed collocation index. In the experiments, the open source search engine Terrier was utilised as a tool to index and retrieve data. The proposed technique was tested on the NTCIR data collection. The experiments demonstrated promising results.

Motivation & Objective

• To improve ad-hoc information retrieval by enhancing query expansion through semantic relatedness.
• To investigate the effectiveness of combining multiple semantic measures—Explicit Semantic Analysis, WordNet path, and collocation index—into a unified relatedness score.
• To evaluate the proposed EWC-based term selection method in a real-world retrieval system using standard benchmarks.
• To demonstrate that integrating diverse semantic signals leads to better retrieval outcomes than individual measures.

Proposed method

• The EWC semantic relatedness measure combines three components: Wikipedia-based Explicit Semantic Analysis, WordNet path-based similarity, and a mixed collocation index.
• Term selection for query expansion is guided by the EWC score, which ranks candidate terms based on semantic relatedness to the original query terms.
• The method leverages pre-indexed Wikipedia and WordNet resources to compute semantic similarity scores efficiently.
• The retrieval system, built on the open-source Terrier engine, indexes and retrieves documents using expanded queries.
• The approach applies a weighted fusion of the three semantic signals to produce a composite relatedness score for each candidate expansion term.
• Experiments use the NTCIR data collection to evaluate performance in standard ad-hoc retrieval scenarios.

Experimental results

Research questions

• RQ1 Can the integration of Wikipedia-based, WordNet-based, and collocation-based semantic signals improve query expansion in ad-hoc retrieval?
• RQ2 How effective is the EWC measure in selecting relevant expansion terms compared to individual semantic measures?
• RQ3 Does the proposed method lead to measurable improvements in standard retrieval metrics on the NTCIR test collection?
• RQ4 What is the impact of combining multiple semantic signals on retrieval effectiveness in a real-world search engine setting?

Key findings

• The EWC-based query expansion method achieved promising results on the NTCIR data collection, demonstrating improved retrieval effectiveness.
• The integration of multiple semantic signals—Explicit Semantic Analysis, WordNet path, and collocation index—yielded better performance than relying on any single measure.
• The proposed method effectively selected relevant expansion terms, leading to enhanced precision and recall in ad-hoc retrieval tasks.
• The experiments confirmed that the EWC measure is a robust and effective approach for semantic term selection in query expansion.
• The use of the Terrier search engine enabled reproducible and scalable evaluation of the method on standard benchmarks.