SelfDR Distills LLM Reasoning Into Direct Recommendation, Cutting Inference Cost

_reachsumit · x · 2026-09-04

SelfDR (CIKM'26) is a self-distillation framework that keeps reasoning's accuracy gains for LLM-based recommendation without paying its inference cost.

• Teacher and student recommenders share the same base LLM; no external models needed
• Teacher: a reasoner trained with downstream recommendation performance as reward generates targeted rationales fed into its input
• Student: a direct recommender learns from the teacher via self-distillation with a dynamic weighting strategy
• Validated on three public datasets for effectiveness, rationality, and efficiency

Code is open-sourced.

Original post →

More from Research

• After Chess and Go: Can Any AI Engine Actually Beat Humans at Scrabble? — zuilserip · 2026-09-20
• Textbook author: 99.9% accuracy can mean zero scientific discoveries — bravo_abad · 2026-09-20
• rasbt: Jev's Secret Sauce Is Data, Not the Algorithm — Laya Rival Falls Far Short — RichmanRonald · 2026-09-20
• Sebastian Raschka open-sources an end-to-end 'AI text detector from scratch' project — rasbt · 2026-09-20
• Sebastian Raschka: restricting LLM outputs to an action space is just a classic encoder classifier — rasbt · 2026-09-20
• Inception Labs CEO Stefano Ermon bets on diffusion LLMs over autoregressive decoding — No Priors · 2026-09-20

Research channel →