TREC_2024_Lateral_Reading_Purpose_and_Need.pptx

Presentation regarding lateral reading concept for encounter misleading information on web

1. TREC 2024 Lateral Reading Track Research by – Presentation by- Dake Zhang Atharva Shinde Mark D. Smucker Pratik Mengal Charles L. A. Clarke (SPPU DOT,Pune ) (University of Waterloo, Canada)

2. What is TREC ? TREC (Text REtrieval Conference) is organized by NIST (U.S. National Institute of Standards and Technology). It provides shared tasks to evaluate search, retrieval, and question-answering systems. The Lateral Reading Track is one such new task from 2024.

3. Problem Statement In the age of misinformation, most people struggle to evaluate online news credibility. The Lateral Reading Track addresses this challenge by promoting systems that think like fact-checkers — cross-verifying information through multiple sources .

4. Purpose and Need Promote research to improve online trust evaluation Develop systems that help users make informed judgments Use question generation and document retrieval for verification Assess how AI (especially LLMs) can assist in identifying trustworthy sources

5. Task Overview Two key task s

6. Methodology Dataset: ClueWeb22-B-English (87 million English webpages) News set: 50 diverse articles from multiple media outlets Evaluation: Human assessors rated quality of questions and usefulness of retrieved documents Metrics: DCG, NDCG, MRR

7. Key Results & Findings Best model ( GPT-4o based) achieved NDCG@10 ≈ 0.73 for Question Generation. Query expansion improved retrieval performance. LLMs help but are inconsistent in question quality. Human-style reasoning remains essential for trust assessment.

8. Future Work Next Step – DRAGUN 2025 DRAGUN = Detection, Retrieval, and Augmented Generation for Understanding News Focus on Retrieval-Augmented Generation (RAG) Aim: Generate attributed reports that explain why an article is trustworthy or not. Encourage broader collaboration and open datasets.

9. Conclusion The TREC Lateral Reading Track bridges human fact-checking techniques and artificial intelligence to make online information evaluation more transparent, efficient, and reliable