Fengyuan Liu, Yi Huang, Sichun Luo, Yuqi Wang, Yazheng Yang, Xinye Li, Zefa Hu, Junlan Feng, Qi Liu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Large language models (LLMs) exhibit varying strengths and weaknesses across different tasks, prompting recent studies to explore the benefits of ensembling models to leverage their complementary advantages. However, existing LLM ensembling methods often overlook model compatibility and struggle with inefficient alignment of probabilities across the entire vocabulary. In this study, we empirically investigate the factors influencing ensemble performance, identifying model performance, vocabulary size, and response style as key determinants, revealing that compatibility among models is essential for effective ensembling. This analysis leads to the development of a simple yet effective model selection strategy that identifies compatible models. Additionally, we introduce the \textsc{Uni}on \textsc{T}op-$k$ \textsc{E}nsembling (\textsc{UniTE}), a novel approach that efficiently combines models by focusing on the union of the top-k tokens from each model, thereby avoiding the need for full vocabulary alignment and reducing computational overhead. Extensive evaluations across multiple benchmarks demonstrate that \textsc{UniTE} significantly enhances performance compared to existing methods, offering a more efficient framework for LLM ensembling.

Large language models (LLMs) have demonstrated remarkable capabilities and have been extensively deployed across various domains, including recommender systems. Prior research has employed specialized prompts to leverage the in-context learning capabilities of LLMs for recommendation purposes. More recent studies have utilized instruction tuning techniques to align LLMs with human preferences, promising more effective recommendations. However, existing methods suffer from several limitations. The full potential of LLMs is not fully elicited due to low-quality tuning data and the overlooked integration of conventional recommender signals. Furthermore, LLMs may generate inconsistent responses for different ranking tasks in the recommendation, potentially leading to unreliable results. In this article, we introduce Ranker for top- k Recommendations (RecRanker), tailored for instruction tuning LLMs to serve as the Ranker for top- k Recommendations. Specifically, we introduce importance-aware sampling, clustering-based sampling, and penalty for repetitive sampling for sampling high-quality, representative, and diverse training data. To enhance the prompt, we introduce a position shifting strategy to mitigate position bias and augment the prompt with auxiliary information from conventional recommendation models, thereby enriching the contextual understanding of the LLM. Subsequently, we utilize the sampled data to assemble an instruction-tuning dataset with the augmented prompts comprising three distinct ranking tasks: pointwise, pairwise, and listwise rankings. We further propose a hybrid ranking method to enhance the model performance by ensembling these ranking tasks. Our empirical evaluations demonstrate the effectiveness of our proposed RecRanker in both direct and sequential recommendation scenarios. 1

Recently, large language models (LLMs) have demonstrated impressive capabilities and gained widespread applications. However, their direct application to recommendation tasks (e.g., rating prediction task) often falls short of optimal results due to a lack of understanding of collaborative information in recommendations. In this paper, we propose Large lAnguage Model Augmented Recommendation (LAMAR) framework to address this limitation. Instead of relying solely on LLMs, our framework combines their outputs with traditional recommendation models, leveraging both collaborative and semantic information. We further enhance the recommendation performance through an ensemble of diverse prompts and utilize LLMs to extract side information for augmenting traditional recommendation models. Empirical studies on real-world datasets demonstrate that LAMAR outperforms existing approaches, highlighting the benefits of leveraging LLMs in recommendation systems. Code is available at https: //github.com/sichunluo/LAMAR.

The recently released GPT-4 Code Interpreter has demonstrated remarkable proficiency in solving challenging math problems, primarily attributed to its ability to seamlessly reason with natural language, generate code, execute code, and continue reasoning based on the execution output. In this paper, we present a method to fine-tune open-source language models, enabling them to use code for modeling and deriving math equations and, consequently, enhancing their mathematical reasoning abilities. We propose a method of generating novel and high-quality datasets with math problems and their code-based solutions, referred to as MathCodeInstruct. Each solution interleaves $\textit{natural language}$, $\textit{code}$, and $\textit{execution results}$. We also introduce a customized supervised fine-tuning and inference approach. This approach yields the MathCoder models, a family of models capable of generating code-based solutions for solving challenging math problems. Impressively, the MathCoder models achieve state-of-the-art scores among open-source LLMs on the MATH (45.2%) and GSM8K (83.9%) datasets, substantially outperforming other open-source alternatives. Notably, the MathCoder model not only surpasses ChatGPT-3.5 and PaLM-2 on GSM8K and MATH but also outperforms GPT-4 on the competition-level MATH dataset. The proposed dataset and models will be released upon acceptance.

Recent progress in large language models (LLMs) like GPT-4 and PaLM-2 has brought significant advancements in addressing math reasoning problems. In particular, OpenAI's latest version of GPT-4, known as GPT-4 Code Interpreter, shows remarkable performance on challenging math datasets. In this paper, we explore the effect of code on enhancing LLMs' reasoning capability by introducing different constraints on the Code Usage Frequency of GPT-4 Code Interpreter. We found that its success can be largely attributed to its powerful skills in generating and executing code, evaluating the output of code execution, and rectifying its solution when receiving unreasonable outputs. Based on this insight, we propose a novel and effective prompting method, explicit $\underline{\text{c}}$ode-based $\underline{\text{s}}$elf-$\underline{\text{v}}$erification (CSV), to further boost the mathematical reasoning potential of GPT-4 Code Interpreter. This method employs a zero-shot prompt on GPT-4 Code Interpreter to encourage it to use code to self-verify its answers. In instances where the verification state registers as "False", the model shall automatically amend its solution, analogous to our approach of rectifying errors during a mathematics examination. Furthermore, we recognize that the states of the verification result indicate the confidence of a solution, which can improve the effectiveness of majority voting. With GPT-4 Code Interpreter and CSV, we achieve an impressive zero-shot accuracy on MATH dataset.

Federated recommendation systems employ federated learning techniques to safeguard user privacy by transmitting model parameters instead of raw user data between user devices and the central server. Nevertheless, the current federated recommender system faces three significant challenges: (1) data heterogeneity: the heterogeneity of users’ attributes and local data necessitates the acquisition of personalized models to improve the performance of federated recommendation; (2) model performance degradation: the privacy-preserving protocol design in the federated recommendation, such as pseudo item labeling and differential privacy, would deteriorate the model performance; (3) communication bottleneck: the standard federated recommendation algorithm can have a high communication overhead. Previous studies have attempted to address these issues, but none have been able to solve them simultaneously. In this article, we propose a novel framework, named PerFedRec++ , to enhance the personalized federated recommendation with self-supervised pre-training. Specifically, we utilize the privacy-preserving mechanism of federated recommender systems to generate two augmented graph views, which are used as contrastive tasks in self-supervised graph learning to pre-train the model. Pre-training enhances the performance of federated models by improving the uniformity of representation learning. Also, by providing a better initial state for federated training, pre-training makes the overall training converge faster, thus alleviating the heavy communication burden. We then construct a collaborative graph to learn the client representation through a federated graph neural network. Based on these learned representations, we cluster users into different user groups and learn personalized models for each cluster. Each user learns a personalized model by combining the global federated model, the cluster-level federated model, and its own fine-tuned local model. Experiments on three real-world datasets show that our proposed method achieves superior performance over existing methods.

The ubiquitous long-tail distribution of inherent user behaviors results in worse recommendation performance for the items with fewer user records (i.e., tail items) than those with richer ones (i.e., head items). Graph-based recommendation methods (e.g., using graph neural networks) have recently emerged as a powerful tool for recommender systems, often outperforming traditional methods. However, existing techniques for alleviating the long-tail problem mainly focus on traditional methods. There is a lack of graph-based methods that can efficiently deal with the long-tail problem.

Federated recommendation applies federated learning techniques in recommendation systems to help protect user privacy by exchanging models instead of raw user data between user devices and the central server. Due to the heterogeneity in user's attributes and local data, attaining personalized models is critical to help improve the federated recommendation performance. In this paper, we propose a Graph Neural Network based Personalized Federated Recommendation (PerFedRec) framework via joint representation learning, user clustering, and model adaptation. Specifically, we construct a collaborative graph and incorporate attribute information to jointly learn the representation through a federated GNN. Based on these learned representations, we cluster users into different user groups and learn personalized models for each cluster. Then each user learns a personalized model by combining the global federated model, the cluster-level federated model, and the user's fine-tuned local model. To alleviate the heavy communication burden, we intelligently select a few representative users (instead of randomly picked users) from each cluster to participate in training. Experiments on real-world datasets show that our proposed method achieves superior performance over existing methods.

The recent popularity of edge devices and Artificial Intelligent of Things (AIoT) has driven a new wave of contextual recommendations, such as location based Point of Interest (PoI) recommendations and computing resource-aware mobile app recommendations. In many such recommendation scenarios, contexts are drifting over time. For example, in a mobile game recommendation, contextual features like locations, battery, and storage levels of mobile devices are frequently drifting over time. However, most existing graph-based collaborative filtering methods are designed under the assumption of static features. Therefore, they would require frequent retraining and/or yield graphical models burgeoning in sizes, impeding their suitability for context-drifting recommendations.