https://github.com/SeongKu-Kang/ToTER_WWW24

Improving Retrieval in Theme-specific Applications using a Corpus Topical Taxonomy (WWW'24)
https://github.com/SeongKu-Kang/ToTER_WWW24

Last synced: about 1 month ago
JSON representation

Improving Retrieval in Theme-specific Applications using a Corpus Topical Taxonomy (WWW'24)

• Host: GitHub
• URL: https://github.com/SeongKu-Kang/ToTER_WWW24
• Owner: SeongKu-Kang
• License: gpl-3.0
• Created: 2024-02-20T10:03:36.000Z (10 months ago)
• Default Branch: main
• Last Pushed: 2024-02-21T06:04:07.000Z (10 months ago)
• Last Synced: 2024-06-04T14:17:52.779Z (7 months ago)
• Language: Jupyter Notebook
• Homepage:
• Size: 27.3 KB
• Stars: 3
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• awesome-taxonomy - https://github.com/SeongKu-Kang/ToTER_WWW24

README

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

This repository provides the source code of "Improving Retrieval in Theme-specific Applications using a Corpus Topical Taxonomy" accepted in TheWebConf (WWW2024) as a research paper.

### 1. Overview

We introduce a new plug-and-play ToTER framework which improves PLM-based retrieval using a corpus topical taxonomy.



    



#### (Training phase) Taxonomy-guided topic class relevance learning

The taxonomy reveals the latent structure of the whole corpus.

To exploit it for retrieval, we first connect the corpus-level knowledge to individual documents.

We formulate this step as an unsupervised multi-label classification, assessing the relevance of each document to each topic class without document-topic labels.

#### (Inference phase) Topical taxonomy-enhanced retrieval

ToTER consists of three strategies to complement the existing retrieve-then-rerank pipeline: (1) search space adjustment, (2) class relevance matching, and (3) query enrichment by core phrases.

Each strategy is designed to gradually focus on fine-grained ranking.

### 2. How to use

Please refer to 'Guide to using ToTER.ipynb' file.

### 3. Resources

- Due to their large size, we provide necessary files (e.g., PLM-embeddings, trained classifier) through another file-sharing system: https://drive.google.com/file/d/1BmUmlAV4i4-lwwQdBkuDsCZS8-OS-y8M/view?usp=sharing