Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Improving Retrieval in Theme-specific Applications using a Corpus Topical Taxonomy

SeongKu Kang, Shivam Agarwal, Bowen Jin, Dongha Lee, Hwanjo Yu, Jiawei Han

TL;DR

The paper addresses retrieval in theme-specific domains where domain terminology and user intents hinder standard PLM-based methods. It introduces ToTER, a framework that leverages a corpus topical taxonomy to align queries and documents through topic class relevance learning, collective topic knowledge distillation, and three inference-time strategies: search space adjustment, class relevance matching, and query enrichment by core phrases. Across academic and product domains, ToTER yields consistent Retrieval+Reranking gains, reduces initial search space via SSA, and improves final rankings through CRM and QEP, even with limited or no labeled data. The approach demonstrates robustness to taxonomy quality and is designed as a plug-and-play enhancement for existing PLM retrievers, offering practical impact for industry-specific search systems.

Abstract

Document retrieval has greatly benefited from the advancements of large-scale pre-trained language models (PLMs). However, their effectiveness is often limited in theme-specific applications for specialized areas or industries, due to unique terminologies, incomplete contexts of user queries, and specialized search intents. To capture the theme-specific information and improve retrieval, we propose to use a corpus topical taxonomy, which outlines the latent topic structure of the corpus while reflecting user-interested aspects. We introduce ToTER (Topical Taxonomy Enhanced Retrieval) framework, which identifies the central topics of queries and documents with the guidance of the taxonomy, and exploits their topical relatedness to supplement missing contexts. As a plug-and-play framework, ToTER can be flexibly employed to enhance various PLM-based retrievers. Through extensive quantitative, ablative, and exploratory experiments on two real-world datasets, we ascertain the benefits of using topical taxonomy for retrieval in theme-specific applications and demonstrate the effectiveness of ToTER.

Improving Retrieval in Theme-specific Applications using a Corpus Topical Taxonomy

TL;DR

The paper addresses retrieval in theme-specific domains where domain terminology and user intents hinder standard PLM-based methods. It introduces ToTER, a framework that leverages a corpus topical taxonomy to align queries and documents through topic class relevance learning, collective topic knowledge distillation, and three inference-time strategies: search space adjustment, class relevance matching, and query enrichment by core phrases. Across academic and product domains, ToTER yields consistent Retrieval+Reranking gains, reduces initial search space via SSA, and improves final rankings through CRM and QEP, even with limited or no labeled data. The approach demonstrates robustness to taxonomy quality and is designed as a plug-and-play enhancement for existing PLM retrievers, offering practical impact for industry-specific search systems.

Abstract

Document retrieval has greatly benefited from the advancements of large-scale pre-trained language models (PLMs). However, their effectiveness is often limited in theme-specific applications for specialized areas or industries, due to unique terminologies, incomplete contexts of user queries, and specialized search intents. To capture the theme-specific information and improve retrieval, we propose to use a corpus topical taxonomy, which outlines the latent topic structure of the corpus while reflecting user-interested aspects. We introduce ToTER (Topical Taxonomy Enhanced Retrieval) framework, which identifies the central topics of queries and documents with the guidance of the taxonomy, and exploits their topical relatedness to supplement missing contexts. As a plug-and-play framework, ToTER can be flexibly employed to enhance various PLM-based retrievers. Through extensive quantitative, ablative, and exploratory experiments on two real-world datasets, we ascertain the benefits of using topical taxonomy for retrieval in theme-specific applications and demonstrate the effectiveness of ToTER.
Paper Structure (36 sections, 7 equations, 3 figures, 12 tables, 1 algorithm)

This paper contains 36 sections, 7 equations, 3 figures, 12 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Problem Formulation
  3. Concept Definition
  4. Problem Definition
  5. Related Work
  6. METHODOLOGY
  7. Topic Class Relevance Learning
  8. Taxonomy-guided silver label generation
  9. Class relevance learning
  10. Training algorithm of ToTER
  11. Topical Taxonomy-Enhanced Retrieval
  12. Search space adjustment (SSA) to reduce search space.
  13. Class relevance matching (CRM) for retriever
  14. Query enrichment by core phrases (QEP) for reranker.
  15. Experiments
  16. ...and 21 more sections

Figures (3)

  • Figure 1: The overview of Topical Taxonomy-Enhanced Retrieval (ToTER) framework.
  • Figure 2: Search space adjustment performance comparison on the academic and product domain.
  • Figure 3: Retrieval results with taxonomy impaired in terms of (left) topic completeness and (right) term coherence.