Training Multi-Vector Models with Inherent Clustering Structure, Reasoning-Enhanced Listwise Document Reranking, and More!

Vol.105 for May 19 - May 25, 2025

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

This week’s newsletter highlights the following research:

1. Training Multi-Vector Models with Inherent Clustering Structure, from Google

2. Learning Sparse Representations for Memory-Efficient Recommender Systems, from Recombee

3. How Weight Decay Encodes Popularity in Recommendation Systems, from Snap Inc.

4. Ultra-Fast Query Processing for LLM-Based Retrieval, from Ma et al.

5. Understanding How LLMs Navigate Internal vs. External Knowledge in Retrieval-Augmented Generation, from Baidu

6. A Comprehensive Analysis of Positional Bias in Modern Information Retrieval Systems, from Zeng et al.

7. Reasoning-Enhanced Listwise Document Reranking, from JHU

8. A Dual-Flow Generative Ranking Network for Scalable Recommendation Systems, from Meituan

9. Identifying and Correcting Mislabeled Negatives in Retrieval Training Sets, from The University of Waterloo

10. Questioning the Value of Reasoning in Passage Reranking, from Jedidi et al.