Yuanchen Bei, Tianxin Wei, Xuying Ning, Yanjun Zhao, Zhining Liu, Xiao Lin, Yada Zhu, Hendrik Hamann, Jingrui He, Hanghang Tong. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Chengming Cui, Tianxin Wei, Ziyi Chen, Ruizhong Qiu, Zhichen Zeng, Zhining Liu, Xuying Ning, Duo Zhou, Jingrui He. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Federated cross-domain recommendation (Federated CDR) aims to collaboratively learn personalized recommendation models across heterogeneous domains while preserving data privacy. Recently, large language model (LLM)-based recommendation models have demonstrated impressive performance by leveraging LLMs' strong reasoning capabilities and broad knowledge. However, adopting LLM-based recommendation models in Federated CDR scenarios introduces new challenges. First, there exists a risk of overfitting with domain-specific local adapters. The magnitudes of locally optimized parameter updates often vary across domains, causing biased aggregation and overfitting toward domain-specific distributions. Second, unlike traditional recommendation models (e.g., collaborative filtering, bipartite graph-based methods) that learn explicit and comparable user/item representations, LLMs encode knowledge implicitly through autoregressive text generation training. This poses additional challenges for effectively measuring the cross-domain similarities under heterogeneity. To address these challenges, we propose an LLM-based framework for federated cross-domain recommendation, FeDecider. Specifically, FeDecider tackles the challenge of scale-specific noise by disentangling each client's low-rank updates and sharing only their directional components. To handle the need for flexible and effective integration, each client further learns personalized weights that achieve the data-aware integration of updates from other domains. Extensive experiments across diverse datasets validate the effectiveness of our proposed FeDecider.

These years have witnessed remarkable progress in modeling user behaviour from personalized online services, especially knowledge graph-based recommendation systems. Meanwhile, more studies are focusing on aspects beyond recommendation performance, since such an observational data-driven paradigm is posing threats to both users and society in terms of trustworthiness. In fact, existing problem-oriented solutions still face significant challenges, as almost all of them suffer from the generality limitations to improve their trustworthiness in a uniform fashion. To address these issues, we propose a plug-and-playDebiasing framework forKnowledgeGraph-basedRecommendationSystems, also known as DiKGRS. Specifically, the Knowledge-augmented Pseudo-Samples Generation (KPSG) method, a novel data augmentation perspective, is proposed to explore more auxiliary information beyond observational user behaviors. Furthermore, the Debiasing Value Networks (DVN), is also developed to evaluate the reliability of generated pseudo-samples by modeling both the item popularity and user demographic bias in the platform. Moreover, an adaptive weighting coordination module is performed to coordinate the proposed DiKGRS framework and its backbones. Experimental results on four real-world datasets from different online service personalization scenarios have illustrated that the proposed framework can significantly improve the trustworthiness of existing knowledge graph-based recommendation systems. The code has been released public available at:https://github.com/alipay/A-Knowledge-augmented-Method-DiKGRS.

Junyu Zhang, Runpei Dong, Han Wang, Xuying Ning, Haoran Geng, Peihao Li, Xialin He, Yutong Bai, Jitendra Malik, Saurabh Gupta, Huan Zhang. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Sequential Recommendation (SR) task involves predicting the next item a user is likely to interact with, given their past interactions. The SR models examine the sequence of a user's actions to discern more complex behavioral patterns and temporal dynamics. Recent research demonstrates the great impact of LLMs on sequential recommendation systems, either viewing sequential recommendation as language modeling or serving as the backbone for user representation. Although these methods deliver outstanding performance, there is scant evidence of the necessity of a large language model and how large the language model is needed, especially in the sequential recommendation scene. Meanwhile, due to the huge size of LLMs, it is inefficient and impractical to apply a LLM-based model in real-world platforms that often need to process billions of traffic logs daily. In this paper, we explore the influence of LLMs' depth by conducting extensive experiments on large-scale industry datasets. Surprisingly, our motivational experiments reveal that most intermediate layers of LLMs are redundant, indicating that pruning the remaining layers can still maintain strong performance. Motivated by this insight, we empower small language models for SR, namely SLMRec, which adopt a simple yet effective knowledge distillation method. Moreover, SLMRec is orthogonal to other post-training efficiency techniques, such as quantization and pruning, so that they can be leveraged in combination. Comprehensive experimental results illustrate that the proposed SLMRec model attains the best performance using only 13\% of the parameters found in LLM-based recommendation models while simultaneously achieving up to 6.6x and 8.0x speedups in training and inference time costs, respectively. Besides, we provide a theoretical justification for why small language models can perform comparably to large language models in SR.

Real-world multimodal data usually exhibit complex structural relationships beyond traditional one-to-one mappings like image-caption pairs. Entities across modalities interact in intricate ways, with images and text forming diverse interconnections through contextual dependencies and co-references. Graphs provide powerful structural information for modeling intra-modal and inter-modal relationships. However, previous works fail to distinguish multi-hop neighbors and treat the graph as a standalone modality, which fragments the overall understanding. This limitation presents two key challenges in multimodal learning: (1) integrating structural information from multi-hop neighbors into foundational models, and (2) fusing modality-specific information in a principled manner. To address these challenges, we revisit the role of graphs in multimodal learning within the era of foundation models and propose Graph4MM, a graph-based multimodal learning framework. To be specific, we introduce Hop-Diffused Attention, which integrates multi-hop structural information into self-attention through causal masking and hop diffusion. Furthermore, we design MM-QFormer, a multi-mapping querying transformer for cross-modal fusion. Through theoretical and empirical analysis, we show that leveraging structures to integrate both intra- and inter-modal interactions improves multimodal understanding beyond treating them as a standalone modality. Experiments on both generative and discriminative tasks show that Graph4MM outperforms larger VLMs, LLMs, and multimodal graph baselines, achieving a 6.93% average improvement.

Cross-Domain Sequential Recommendation (CDSR) leverages user behaviors across multiple domains to mitigate data sparsity and cold-start challenges in Single-Domain Sequential Recommendation. Existing methods primarily rely on shared users (overlapping users) to learn transferable interest representations. However, these approaches have limited information propagation, benefiting mainly overlapping users and those with rich interaction histories while neglecting non-overlapping (cold-start) and long-tailed users, who constitute the majority in real-world scenarios. To address this issue, we propose i$^2$VAE, a novel variational autoencoder (VAE)-based framework that enhances user interest learning with mutual information-based regularizers. i$^2$VAE improves recommendations for cold-start and long-tailed users while maintaining strong performance across all user groups. Specifically, cross-domain and disentangling regularizers extract transferable features for cold-start users, while a pseudo-sequence generator synthesizes interactions for long-tailed users, refined by a denoising regularizer to filter noise and preserve meaningful interest signals. Extensive experiments demonstrate that i$^2$VAE outperforms state-of-the-art methods, underscoring its effectiveness in real-world CDSR applications. Code and datasets are available at https://github.com/WujiangXu/IM-VAE.