Implicit Reasoning for LLM-based Generative Recommendation, Why Stronger Encoders Make Weaker SPLADE Models, and More!
Vol.161 for Jun 15 - Jun 21, 2026
Stay Ahead of the Curve with the Latest Advancements and Discoveries in Information Retrieval.
This week’s newsletter highlights the following research:
Implicit Reasoning over Semantic IDs for LLM Recommenders, from Snap Inc
Calibrating MLM-Head Scale for Learned Sparse Retrieval, from Korea University
Compressing Diverse Signals into Soft Tokens for Large Recommendation Models, from Google
Offline Indexing-Time Reasoning for Reasoning-Intensive Retrieval, from Lei et al.
The Memorization Trap in LLM-Based Generative Recommendation, from Snap Inc
Building Deep Research Agents from Verifiable Agentic Trajectories, from Wenge AI
Content-Guided Denoising of Implicit Feedback for Cold-Start Recommendation, from Kuaishou
A Taxonomy and Empirical Study of Failure Modes in N-Gram Generative Retrieval, from Vienna University of Economics and Business
Measuring and Fixing Evidence Dilution in Long-Document Dense Retrieval, from Lyu et al.
Diagnosing and Fixing Redundancy in Parallel Agentic Search, from Murali et al.
