arXiv AI

FAIDER: Fairness-Aware Item-Level Decoding for Large Language Model Recommenders

FAIDER is a fairness-aware decoding method for large language model recommenders that selects complete item candidates instead of manipulating token-level logits, thereby reducing semantic disparity across protected attribute groups. It uses conformal-inspired calibration to set a cross-group disparity threshold and applies either hard filtering or a soft penalty during online decoding, with dual-inspired feedback to target a specific violation rate. The approach offers a theoretical guarantee that the empirical violation rate of the soft decoder exceeds its target by at most $O(1/T)$ and demonstrates improved semantic fairness and ranking quality over prompt-repair baselines in experiments.

arXiv AI
Sep 15

Safety as a Constraint: Fine-Tuning a LLM Recommender to Explain Itself

The paper presents a method for fine‑tuning a large language model (LLM) recommender to generate personalized, non‑harmful explanations for its recommendations. By training two LLM‑judge reward models and using constrained GRPO, the authors achieve a significant increase in the PASS rate for all three criteria, from 0.649 to 0.956 on their own judges and from 0.677 to 0.931 on an independent judge. The fine‑tuned model maintains its original recommendation performance, demonstrating that LLM‑based recommenders can be adapted to complex tasks without loss of effectiveness.

By Jiashu He, Emma Yanyang Kong, JJ Tan, David Fagnan
arXiv Machine Learning
Sep 3

GenCAR: Generative Counterfactual Alignment with Risk-Controlled Selection for Out-of-Distribution Recommendation

GenCAR introduces a method for out‑of‑distribution recommendation that balances utility and risk by controlling the proxy‑label false discovery rate (FDR). It frames the problem as an α‑Valid Counterfactual Recommendation (α‑VCR) task, coupling counterfactual supervision with calibrated set selection using conformal p‑values and Benjamini–Hochberg filtering. The approach theoretically bounds counterfactual approximation error and guarantees finite‑sample, distribution‑free FDR control under various dependence assumptions, and empirical results show improved OOD candidate recovery across benchmarks.

By Qianqian Wang, Yunshan Li, Jiawen Zeng, Wenwu Gong, Lili Yang
arXiv AI
Sep 2

Retrieval, Scoring, and Decoding Shape Performance and Stability in LLM-based Conversational Recommendation

The study evaluates large language models (LLMs) as rerankers in conversational movie recommendation, comparing proprietary, open-weight, and fine-tuned LLMs against collaborative-filtering and sequential baselines on the ReDial benchmark. Results show that the best proprietary LLM achieves an NDCG@10 of 0.1497 with a shared semantic candidate pool, outperforming non-LLM baselines, while open-weight LLMs do not surpass a tuned shallow autoencoder under the same protocol. The analysis also highlights that reranker performance is highly sensitive to candidate generation, pool size, scoring policy, and decoding temperature, suggesting these factors should be reported as standard evaluation fields.

By Ante Kapetanovic, Tomislav Duricic, Andro Mercep, Emanuel Lacic
arXiv Computation and Language
Nov 4

Preprocessing Choices in Fine-Tuning Data Matter: Utility and Demographic Sensitivity in LLM-Based Recommendation

The paper examines how different preprocessing steps in fine‑tuning data influence the performance and fairness of a generative large language model recommender system. Using a controlled pipeline that includes supervised fine‑tuning, catalog‑constrained decoding, and prompt augmentation, the authors evaluate three preprocessing choices—item‑title representation, demographic balancing, and auxiliary item‑knowledge enrichment—on the MovieLens‑1M dataset. Results show that catalog constraints eliminate out‑of‑catalog items, prompt augmentation improves consistency and utility, and that demographic balancing and knowledge enrichment can reduce utility gaps for certain groups while preserving overall performance, though gender balancing may lower utility without improving gender indicators.

By Furkan Yasir Goksu, Yaren Yilmaz, Sule Gunduz Oguducu
arXiv AI
6d ago

From Counterfactual Edits to Human Narratives: LLM-Driven Interpretations for Recommender System Explanations

The paper explores how large language models can transform extensive counterfactual edit lists from recommender systems into concise, natural‑language explanations that highlight higher‑level preference patterns. By optimizing these mediators for correctness, informativeness, and linguistic quality, the authors demonstrate that the resulting summaries maintain high linguistic quality and improve semantic coherence, especially when paired with effective counterfactual generators like ACCENT and SPINRec. Human‑calibrated metrics confirm that preference optimization enhances the overall quality of the explanations across multiple settings.

By Vojtech Cahlik, Amir Reza Mohammadi, Rodrigo Alves, Pavel Kordik, Eva Zangerle