Extremely Long User History Modeling at Instagram, When Compact Embeddings Outperform Large Models in Hybrid RAG, and More!

Vol.107 for Jun 02 - Jun 08, 2025

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

This week’s newsletter highlights the following research:

1. When Compact Embeddings Outperform Large Models in Hybrid RAG, from SAP

2. Evaluating Multi-Stage vs. Simple Retrieval Strategies for Long-Context Language Models, from Stanford University

3. Practical Token Reduction for Large-Scale Visual Document Retrieval, from Ma et al.

4. Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models, from Alibaba

5. Scaling User History Length in Production Recommendation Systems, from Meta

6. An Agentic Approach to Multimodal Research Report Generation, from Zhejiang University

7. Learning Step-by-Step Reasoning-Retrieval Strategies for Enhanced Question Answering, from Fudan University

8. Maintaining Dense Retriever Performance in Dynamic Corpora via Gradient-Based OOD Detection, from Ko et al.

9. A Two-Stage Training Framework for Efficient Small Language Model Reranking, from Mixedbread AI

10. Scalable Intent Classification Through Retrieval-Augmented Generation, from Amazon