Zhu Sun

Zhijun Chen, Zeyu Ji, Qianren Mao et al.

We propose LLM-PeerReview, an unsupervised LLM Ensemble method that selects the most ideal response from multiple LLM-generated candidates for each query, harnessing the collective wisdom of multiple models with diverse strengths. LLM-PeerReview is built on a novel, peer-review-inspired framework that offers a transparent and interpretable mechanism, while remaining fully unsupervised for flexible adaptability and generalization. Specifically, it operates in three stages: For scoring, we use the emerging LLM-as-a-Judge technique to evaluate each response by reusing multiple LLMs at hand; For reasoning, we can apply a straightforward averaging strategy or a principled graphical model-based truth inference algorithm to aggregate multiple scores to produce a final score for each response; Finally, the highest-scoring response is selected as the best ensemble output. LLM-PeerReview is conceptually simple and empirically powerful. Our results across four datasets show that the two variants of the proposed approach outperform the advanced model Smoothie-Global by 6.9% and 7.3% points, cross diverse task types including factual recall QA, math reasoning, and instruction following.

Kaidong Feng, Zhu Sun, Jie Yang et al.

LLMs are increasingly explored for bundle generation, thanks to their reasoning capabilities and knowledge. However, deploying large-scale LLMs introduces significant efficiency challenges, primarily high computational costs during fine-tuning and inference due to their massive parameterization. Knowledge distillation (KD) offers a promising solution, transferring expertise from large teacher models to compact student models. This study systematically investigates knowledge distillation approaches for bundle generation, aiming to minimize computational demands while preserving performance. We explore three critical research questions: (1) how does the format of KD impact bundle generation performance? (2) to what extent does the quantity of distilled knowledge influence performance? and (3) how do different ways of utilizing the distilled knowledge affect performance? We propose a comprehensive KD framework that (i) progressively extracts knowledge (patterns, rules, deep thoughts); (ii) captures varying quantities of distilled knowledge through different strategies; and (iii) exploits complementary LLM adaptation techniques (in-context learning, supervised fine-tuning, combination) to leverage distilled knowledge in small student models for domain-specific adaptation and enhanced efficiency. Extensive experiments provide valuable insights into how knowledge format, quantity, and utilization methodologies collectively shape LLM-based bundle generation performance, exhibiting KD's significant potential for more efficient yet effective LLM-based bundle generation.

Qing Yin, Hui Fang, Zhu Sun et al.

Session-based recommender systems (SBRSs) have become extremely popular in view of the core capability of capturing short-term and dynamic user preferences. However, most SBRSs primarily maximize recommendation accuracy but ignore user minor preferences, thus leading to filter bubbles in the long run. Only a handful of works, being devoted to improving diversity, depend on unique model designs and calibrated loss functions, which cannot be easily adapted to existing accuracy-oriented SBRSs. It is thus worthwhile to come up with a simple yet effective design that can be used as a plugin to facilitate existing SBRSs on generating a more diversified list in the meantime preserving the recommendation accuracy. In this case, we propose an end-to-end framework applied for every existing representative (accuracy-oriented) SBRS, called diversified category-aware attentive SBRS (DCA-SBRS), to boost the performance on recommendation diversity. It consists of two novel designs: a model-agnostic diversity-oriented loss function, and a non-invasive category-aware attention mechanism. Extensive experiments on three datasets showcase that our framework helps existing SBRSs achieve extraordinary performance in terms of recommendation diversity and comprehensive performance, without significantly deteriorating recommendation accuracy compared to state-of-the-art accuracy-oriented SBRSs.