Training Retrievers for Reasoning Tasks, A Unified Framework for Text, Image, and Video RAG, and More!

Vol.102 for Apr 28 - May 04, 2025

Stay Ahead of the Curve with the Latest Advancements and Discoveries in Information Retrieval.

This week’s newsletter highlights the following research:

1. A Multi-Modal, Multi-Granular Framework for Adaptive Knowledge Retrieval, from KAIST

2. Training Models for Reasoning-Intensive Information Retrieval, from Meta

3. Comparative Performance of Advanced Chunking Techniques in Retrieval-Augmented Generation, from the University of Bologna

4. Guardrails for Valid LLM-Based Evaluation in Information Retrieval, from Dietz et al.

5. Learning Universal User Representations for Multi-Surface Recommendation Systems, from Snap Inc.

6. Efficient Top-K Recommendation in Million-Item Catalogs Without Exhaustive Scoring, from Petrov et al.

7. A Dual-Process Framework for Dynamic Multi-Hop Question Answering, from Cheng et al.

8. Revealing the Limitations of Hallucination Detection Metrics, from Apple

9. Multi-Task Contrastive Learning for Universal Embeddings in Billion-Node Graphs, from Pinterest

10. Learning Sequential Recommendation Embeddings in the Null Space of Language Models, from Hu et al.