Qiaoyu Tan
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Qiaoyu Tan has a total of 106 co-authors. Here are the top co-authors by number of shared publications:
- Xiao Huang
The Hong Kong Polytechnic University, Hong Kong, Hong Kong
9 shared publications - Ninghao Liu
The Hong Kong Polytechnic University, Hong Kong, Hong Kong
7 shared publications - Xia Hu
Rice University, Houston, TX, USA
5 shared publications - Yuanfu Sun
Department of Computer Science New York University ( Shanghai
5 shared publications - Daochen Zha
Airbnb, San Francisco, USA
3 shared publications - Jiajun Bu
Zhejiang Key Lab of Accessible Perception & Intelligent Systems, Zhejiang University, Hangzhou, China
3 shared publications
Latest Publications
- Open Access
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 1027-1038
MLaGA: Multimodal Large Language and Graph Assistant
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 1027-1038
Open AccessMLaGA: Multimodal Large Language and Graph Assistant
- Open Access
- 21 April 2026
- Published by Springer Nature
- p. 189-208
Recommender Systems for STEM Education Research
- 21 April 2026
- Published by Springer Nature
- p. 189-208
Open AccessRecommender Systems for STEM Education Research
- Open Access
- 25 March 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Information Systems
- Vol. 44  (3) , 1-40
Federated Recommendation via Stochastic Aggregation and Consistency Inference
With growing concerns over user privacy, federated recommendation (FedRec) has emerged as a mainstream solution for personalized recommendation services. FedRec trains user-private parameters on local clients while collaboratively updating global parameters on a centralized server. However, despite advances in optimizing these local and global parameters, existing methods overlook two key challenges: tradeoff training and distribution discrepancy . Tradeoff training balances timely local updates with diverse global parameters, limiting the model’s learning ability. Distribution discrepancy arises from the divergence between locally trained global parameters and those aggregated by the server, corrupting inference performance. To fill in the gap, we propose FedSC , a principled federated recommendation framework that boosts FedRec’s training and inference processes with minimal yet nontrivial efforts. During training, FedSC employs a stochastic aggregation strategy where all users participate in every round, while only a random subset is selected for aggregation, preserving the diversity of global parameters and ensuring timely local updates. During inference, FedSC makes recommendations with a consistency inference mechanism that uses the most recent locally trained global parameters of each user to improve the model’s understanding of user preferences. Extensive experiments on multiple benchmark datasets demonstrate the superiority of FedSC, achieving up to a 20% improvement in most evaluation scenarios.
- 25 March 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Information Systems
- Vol. 44  (3) , 1-40
Open AccessFederated Recommendation via Stochastic Aggregation and Consistency Inference
With growing concerns over user privacy, federated recommendation (FedRec) has emerged as a mainstream solution for personalized recommendation services. FedRec trains user-private parameters on local clients while collaboratively updating global parameters on a centralized server. However, despite advances in optimizing these local and global parameters, existing methods overlook two key challenges: tradeoff training and distribution discrepancy . Tradeoff training balances timely local updates with diverse global parameters, limiting the model’s learning ability. Distribution discrepancy arises from the divergence between locally trained global parameters and those aggregated by the server, corrupting inference performance. To fill in the gap, we propose FedSC , a principled federated recommendation framework that boosts FedRec’s training and inference processes with minimal yet nontrivial efforts. During training, FedSC employs a stochastic aggregation strategy where all users participate in every round, while only a random subset is selected for aggregation, preserving the diversity of global parameters and ensuring timely local updates. During inference, FedSC makes recommendations with a consistency inference mechanism that uses the most recent locally trained global parameters of each user to improve the model’s understanding of user preferences. Extensive experiments on multiple benchmark datasets demonstrate the superiority of FedSC, achieving up to a 20% improvement in most evaluation scenarios.
- Open Access
GraphSearch: Agentic Search-Augmented Reasoning for Zero-Shot Graph Learning
Recent advances in search-augmented large reasoning models (LRMs) enable the retrieval of external knowledge to reduce hallucinations in multistep reasoning. However, their ability to operate on graph-structured data, prevalent in domains such as e-commerce, social networks, and scientific citations, remains underexplored. Unlike plain text corpora, graphs encode rich topological signals that connect related entities and can serve as valuable priors for retrieval, enabling more targeted search and improved reasoning efficiency. Yet, effectively leveraging such structure poses unique challenges, including the difficulty of generating graph-expressive queries and ensuring reliable retrieval that balances structural and semantic relevance. To address this gap, we introduce GraphSearch, the first framework that extends search-augmented reasoning to graph learning, enabling zero-shot graph learning without task-specific fine-tuning. GraphSearch combines a Graph-aware Query Planner, which disentangles search space (e.g., 1-hop, multi-hop, or global neighbors) from semantic queries, with a Graph-aware Retriever, which constructs candidate sets based on topology and ranks them using a hybrid scoring function. We further instantiate two traversal modes: GraphSearch-R, which recursively expands neighborhoods hop by hop, and GraphSearch-F, which flexibly retrieves across local and global neighborhoods without hop constraints. Extensive experiments across diverse benchmarks show that GraphSearch achieves competitive or even superior performance compared to supervised graph learning methods, setting state-of-the-art results in zero-shot node classification and link prediction. These findings position GraphSearch as a flexible and generalizable paradigm for agentic reasoning over graphs.
Open AccessGraphSearch: Agentic Search-Augmented Reasoning for Zero-Shot Graph Learning
Recent advances in search-augmented large reasoning models (LRMs) enable the retrieval of external knowledge to reduce hallucinations in multistep reasoning. However, their ability to operate on graph-structured data, prevalent in domains such as e-commerce, social networks, and scientific citations, remains underexplored. Unlike plain text corpora, graphs encode rich topological signals that connect related entities and can serve as valuable priors for retrieval, enabling more targeted search and improved reasoning efficiency. Yet, effectively leveraging such structure poses unique challenges, including the difficulty of generating graph-expressive queries and ensuring reliable retrieval that balances structural and semantic relevance. To address this gap, we introduce GraphSearch, the first framework that extends search-augmented reasoning to graph learning, enabling zero-shot graph learning without task-specific fine-tuning. GraphSearch combines a Graph-aware Query Planner, which disentangles search space (e.g., 1-hop, multi-hop, or global neighbors) from semantic queries, with a Graph-aware Retriever, which constructs candidate sets based on topology and ranks them using a hybrid scoring function. We further instantiate two traversal modes: GraphSearch-R, which recursively expands neighborhoods hop by hop, and GraphSearch-F, which flexibly retrieves across local and global neighborhoods without hop constraints. Extensive experiments across diverse benchmarks show that GraphSearch achieves competitive or even superior performance compared to supervised graph learning methods, setting state-of-the-art results in zero-shot node classification and link prediction. These findings position GraphSearch as a flexible and generalizable paradigm for agentic reasoning over graphs.
- 7 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Knowledge and Data Engineering
- Vol. 38  (2) , 1028-1039
Enhancing Graph Condensation via Key Information Reconstruction
Graph data mining techniques in real-world scenarios often encounter significant computational challenges, especially when the graph contains a large number of nodes and edges. Recently, Graph Condensation (GC) has emerged to offer data-centric solutions that address the challenge of graph volume, enhancing the efficiency of graph data mining and storage. Current methods in GC rely solely on optimizing heuristic metrics of one-way maintenance of key information in the condensed graph. However, the maintenance of key information may be insufficient due to the significant condensation ratio, yet these methods lack an effective mechanism to verify and compensate for that. To this end, this paper aims to enhance the maintenance of key information through a reconstruction-based alignment mechanism. More specifically, inspired by the Kolmogorov Complexity, we revisit the theoretical foundations of GC and propose a way-back mechanism that introduces a feedback loop of learning to reconstruct the original graph from the condensed graph, with the objective of key information alignment, namely the WbGC. We modify several GC methods with our mechanism, and the experiments show that our approach provides an enhanced solution for GC.
- 7 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Knowledge and Data Engineering
- Vol. 38  (2) , 1028-1039
Enhancing Graph Condensation via Key Information Reconstruction
Graph data mining techniques in real-world scenarios often encounter significant computational challenges, especially when the graph contains a large number of nodes and edges. Recently, Graph Condensation (GC) has emerged to offer data-centric solutions that address the challenge of graph volume, enhancing the efficiency of graph data mining and storage. Current methods in GC rely solely on optimizing heuristic metrics of one-way maintenance of key information in the condensed graph. However, the maintenance of key information may be insufficient due to the significant condensation ratio, yet these methods lack an effective mechanism to verify and compensate for that. To this end, this paper aims to enhance the maintenance of key information through a reconstruction-based alignment mechanism. More specifically, inspired by the Kolmogorov Complexity, we revisit the theoretical foundations of GC and propose a way-back mechanism that introduces a feedback loop of learning to reconstruct the original graph from the condensed graph, with the objective of key information alignment, namely the WbGC. We modify several GC methods with our mechanism, and the experiments show that our approach provides an enhanced solution for GC.
- 23 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Knowledge and Data Engineering
- Vol. 37  (12) , 7016-7029
Cross-Task Collaborative Meta-Learning for Cold-Start Recommendations
Optimizer-based meta-learning, specifically model-agnostic meta-learning (MAML), has emerged as a powerful tool for tackling the cold-start recommendation problem. In these meta-learning-based methods, recommendations for individual users are typically treated as separate tasks and learned independently. However, this task-by-task learning paradigm presents several observable limitations. First, learning one task at a time ignores inter-task correlations, i.e., collaborative signals, which limits the meta-model’s receptive field and prevents it from leveraging valuable shared information, ultimately leading to subpar performance. Second, the meta-model is susceptible to the task distribution, i.e., the varied preference distributions among different users, which in turn introduces biases and inconsistencies, resulting in a less robust model that may perform well on certain user groups while underperforming on others. In this paper, we explore the correlations among different tasks in cold-start recommendations and develop a novel strategy termed cross-task collaborative meta-learning (CCML). More specifically, we propose a collaborative task sampling module designed to mitigate the adverse impact of irrelevant tasks during meta-model learning. This module adaptively identifies tasks that are both similar and beneficial to the primary task, ensuring that the meta-model learns from relevant and supportive information. Additionally, to harness collaborative information across relevant tasks, we introduce a bi-level cross-task meta-training strategy. This strategy leverages multi-task learning to capture collaborative knowledge simultaneously and enhance user profiling with pertinent information. Extensive experiments on four public benchmark datasets demonstrate the advantages of CCML over many state-of-the-art cold-start recommendation methods. Our results show significant improvements in recommendation accuracy and robustness, highlighting the potential of cross-task collaboration in enhancing meta-learning-based recommender systems.
- 23 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Knowledge and Data Engineering
- Vol. 37  (12) , 7016-7029
Cross-Task Collaborative Meta-Learning for Cold-Start Recommendations
Optimizer-based meta-learning, specifically model-agnostic meta-learning (MAML), has emerged as a powerful tool for tackling the cold-start recommendation problem. In these meta-learning-based methods, recommendations for individual users are typically treated as separate tasks and learned independently. However, this task-by-task learning paradigm presents several observable limitations. First, learning one task at a time ignores inter-task correlations, i.e., collaborative signals, which limits the meta-model’s receptive field and prevents it from leveraging valuable shared information, ultimately leading to subpar performance. Second, the meta-model is susceptible to the task distribution, i.e., the varied preference distributions among different users, which in turn introduces biases and inconsistencies, resulting in a less robust model that may perform well on certain user groups while underperforming on others. In this paper, we explore the correlations among different tasks in cold-start recommendations and develop a novel strategy termed cross-task collaborative meta-learning (CCML). More specifically, we propose a collaborative task sampling module designed to mitigate the adverse impact of irrelevant tasks during meta-model learning. This module adaptively identifies tasks that are both similar and beneficial to the primary task, ensuring that the meta-model learns from relevant and supportive information. Additionally, to harness collaborative information across relevant tasks, we introduce a bi-level cross-task meta-training strategy. This strategy leverages multi-task learning to capture collaborative knowledge simultaneously and enhance user profiling with pertinent information. Extensive experiments on four public benchmark datasets demonstrate the advantages of CCML over many state-of-the-art cold-start recommendation methods. Our results show significant improvements in recommendation accuracy and robustness, highlighting the potential of cross-task collaboration in enhancing meta-learning-based recommender systems.
- Open Access
CrystalICL: Enabling In-Context Learning for Crystal Generation
Designing crystal materials with desired physicochemical properties remains a fundamental challenge in materials science. While large language models (LLMs) have demonstrated strong in-context learning (ICL) capabilities, existing LLM-based crystal generation approaches are limited to zero-shot scenarios and are unable to benefit from few-shot scenarios. In contrast, human experts typically design new materials by modifying relevant known structures which aligns closely with the few-shot ICL paradigm. Motivated by this, we propose CrystalICL, a novel model designed for few-shot crystal generation. Specifically, we introduce a space-group based crystal tokenization method, which effectively reduces the complexity of modeling crystal symmetry in LLMs. We further introduce a condition-structure aware hybrid instruction tuning framework and a multi-task instruction tuning strategy, enabling the model to better exploit ICL by capturing structure-property relationships from limited data. Extensive experiments on four crystal generation benchmarks demonstrate the superiority of CrystalICL over the leading baseline methods on conditional and unconditional generation tasks.
Open AccessCrystalICL: Enabling In-Context Learning for Crystal Generation
Designing crystal materials with desired physicochemical properties remains a fundamental challenge in materials science. While large language models (LLMs) have demonstrated strong in-context learning (ICL) capabilities, existing LLM-based crystal generation approaches are limited to zero-shot scenarios and are unable to benefit from few-shot scenarios. In contrast, human experts typically design new materials by modifying relevant known structures which aligns closely with the few-shot ICL paradigm. Motivated by this, we propose CrystalICL, a novel model designed for few-shot crystal generation. Specifically, we introduce a space-group based crystal tokenization method, which effectively reduces the complexity of modeling crystal symmetry in LLMs. We further introduce a condition-structure aware hybrid instruction tuning framework and a multi-task instruction tuning strategy, enabling the model to better exploit ICL by capturing structure-property relationships from limited data. Extensive experiments on four crystal generation benchmarks demonstrate the superiority of CrystalICL over the leading baseline methods on conditional and unconditional generation tasks.
- Open Access
- 3 August 2025
- Published by Association for Computing Machinery (ACM)
- p. 5912-5923
TrustGLM: Evaluating the Robustness of GraphLLMs Against Prompt, Text, and Structure Attacks
- 3 August 2025
- Published by Association for Computing Machinery (ACM)
- p. 5912-5923
Open AccessTrustGLM: Evaluating the Robustness of GraphLLMs Against Prompt, Text, and Structure Attacks
- Open Access
- 20 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1913-1923
DimCL: Dimension-Aware Augmentation in Contrastive Learning for Recommendation
- 20 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1913-1923
Open AccessDimCL: Dimension-Aware Augmentation in Contrastive Learning for Recommendation
- Open Access
Graph-MLLM: Harnessing Multimodal Large Language Models for Multimodal Graph Learning
Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities in representing and understanding diverse modalities. However, they typically focus on modality alignment in a pairwise manner while overlooking structural relationships across data points. Integrating multimodality with structured graph information (i.e., multimodal graphs, MMGs) is essential for real-world applications such as social networks, healthcare, and recommendation systems. Existing MMG learning methods fall into three paradigms based on how they leverage MLLMs: Encoder, Aligner, and Predictor. MLLM-as-Encoder focuses on enhancing graph neural networks (GNNs) via multimodal feature fusion; MLLM-as-Aligner aligns multimodal attributes in language or hidden space to enable LLM-based graph reasoning; MLLM-as-Predictor treats MLLMs as standalone reasoners with in-context learning or fine-tuning. Despite their advances, the MMG field lacks a unified benchmark to fairly evaluate across these approaches, making it unclear what progress has been made. To bridge this gap, we present Graph-MLLM, a comprehensive benchmark for multimodal graph learning by systematically evaluating these three paradigms across six datasets with different domains. Through extensive experiments, we observe that jointly considering the visual and textual attributes of the nodes benefits graph learning, even when using pre-trained text-to-image alignment models (e.g., CLIP) as encoders. We also find that converting visual attributes into textual descriptions further improves performance compared to directly using visual inputs. Moreover, we observe that fine-tuning MLLMs on specific MMGs can achieve state-of-the-art results in most scenarios, even without explicit graph structure information. We hope that our open-sourced library will facilitate rapid, equitable evaluation and inspire further innovative research in this field.
Open AccessGraph-MLLM: Harnessing Multimodal Large Language Models for Multimodal Graph Learning
Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities in representing and understanding diverse modalities. However, they typically focus on modality alignment in a pairwise manner while overlooking structural relationships across data points. Integrating multimodality with structured graph information (i.e., multimodal graphs, MMGs) is essential for real-world applications such as social networks, healthcare, and recommendation systems. Existing MMG learning methods fall into three paradigms based on how they leverage MLLMs: Encoder, Aligner, and Predictor. MLLM-as-Encoder focuses on enhancing graph neural networks (GNNs) via multimodal feature fusion; MLLM-as-Aligner aligns multimodal attributes in language or hidden space to enable LLM-based graph reasoning; MLLM-as-Predictor treats MLLMs as standalone reasoners with in-context learning or fine-tuning. Despite their advances, the MMG field lacks a unified benchmark to fairly evaluate across these approaches, making it unclear what progress has been made. To bridge this gap, we present Graph-MLLM, a comprehensive benchmark for multimodal graph learning by systematically evaluating these three paradigms across six datasets with different domains. Through extensive experiments, we observe that jointly considering the visual and textual attributes of the nodes benefits graph learning, even when using pre-trained text-to-image alignment models (e.g., CLIP) as encoders. We also find that converting visual attributes into textual descriptions further improves performance compared to directly using visual inputs. Moreover, we observe that fine-tuning MLLMs on specific MMGs can achieve state-of-the-art results in most scenarios, even without explicit graph structure information. We hope that our open-sourced library will facilitate rapid, equitable evaluation and inspire further innovative research in this field.