Kai Han
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Kai Han has a total of 74 co-authors. Here are the top co-authors by number of shared publications:
- Jing Tang
The Hong Kong University of Science and Technology
10 shared publications - Shuang Cui
Soochow University, Suzhou, Jiangsu, China
10 shared publications - He Huang
Soochow University, Suzhou, China
9 shared publications - Jun Luo
Nanyang Technological University
7 shared publications - Shaojie Tang
Department of Management Science and Systems, University at Buffalo, Buffalo, New York, USA
6 shared publications - Zhizhuo Yin
Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China
5 shared publications
Latest Publications
- Open Access
- 19 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1644-1655
Mitigating Structural Overfitting: A Distribution-Aware Rectification Framework for Missing Feature Imputation
Incomplete node features are ubiquitous in real-world scenarios such as user profiling and cold-start recommendation, which severely hinders the practical deployment of graph learning systems (e.g., GNNs). Existing solutions typically rely on diffusion-based structural smoothing (e.g., feature propagation) to impute missing values. However, we find that these approaches suffer from structural overfitting, leading to three progressive challenges: 1) performance degradation on disjoint graphs, 2) loss of semantic diversity due to over-smoothing, and 3) feature distribution shift when generalizing to unseen graph structures (inductive tasks). To address these challenges, we introduce the DART framework. It begins by employing Global Structural Augmentation (GSA), which establishes global correlations to bridge disjoint components and extend diffusion coverage. Building upon this, we design a semantic rectifier based on masked autoencoding. This module learns the latent feature manifold to recover natural semantic details. Crucially, we introduce a test-time distribution rectification mechanism that projects structurally biased features back onto the learned manifold during inference, effectively bridging the inductive distribution gap. Furthermore, considering that synthetic masking fails to reflect realworld sparsity, we present a new dataset Sailing collected from voyage records with naturally missing attributes. Extensive experiments on six public datasets and Sailing demonstrate that DART significantly outperforms state-of-the-art methods in both transductive and inductive settings. Our code and dataset are available at https://github.com/yfsong00/DART.
- 19 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1644-1655
Open AccessMitigating Structural Overfitting: A Distribution-Aware Rectification Framework for Missing Feature Imputation
Incomplete node features are ubiquitous in real-world scenarios such as user profiling and cold-start recommendation, which severely hinders the practical deployment of graph learning systems (e.g., GNNs). Existing solutions typically rely on diffusion-based structural smoothing (e.g., feature propagation) to impute missing values. However, we find that these approaches suffer from structural overfitting, leading to three progressive challenges: 1) performance degradation on disjoint graphs, 2) loss of semantic diversity due to over-smoothing, and 3) feature distribution shift when generalizing to unseen graph structures (inductive tasks). To address these challenges, we introduce the DART framework. It begins by employing Global Structural Augmentation (GSA), which establishes global correlations to bridge disjoint components and extend diffusion coverage. Building upon this, we design a semantic rectifier based on masked autoencoding. This module learns the latent feature manifold to recover natural semantic details. Crucially, we introduce a test-time distribution rectification mechanism that projects structurally biased features back onto the learned manifold during inference, effectively bridging the inductive distribution gap. Furthermore, considering that synthetic masking fails to reflect realworld sparsity, we present a new dataset Sailing collected from voyage records with naturally missing attributes. Extensive experiments on six public datasets and Sailing demonstrate that DART significantly outperforms state-of-the-art methods in both transductive and inductive settings. Our code and dataset are available at https://github.com/yfsong00/DART.
- 1 May 2026
- Published by Springer Nature
- p. 21-38
Improved Budget-Feasible Mechanisms for Submodular Valuations: Beating 4 Deterministically in Linear Time
- 1 May 2026
- Published by Springer Nature
- p. 21-38
Improved Budget-Feasible Mechanisms for Submodular Valuations: Beating 4 Deterministically in Linear Time
- Open Access
- 20 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1904-1915
IMGNN: An Efficient, Effective and Generalizable Algorithm for Influence Maximization in Social Networks
Influence Maximization (IM) is a crucial problem in social network analysis and has been extensively studied. Traditional approaches rely on designing approximation algorithms using network sampling; however, these methods lack generalizability and depend on an explicit definition of the influence diffusion model as input. Recently, researchers have turned to deep learning methods to address the shortcomings of traditional IM algorithms, but current learning-based IM algorithms still suffer from severe deficiencies in scalability and generalizability. In this paper, we propose IMGNN, a simple, efficient, effective, and generalizable algorithm powered by graph neural networks. IMGNN is a learning-based IM algorithm with strong generalization capability that reduces the overhead of model retraining while also providing fast seed set inference speed. As a result, IMGNN achieves better performance in terms of both efficiency and effectiveness compared to existing IM algorithms. Its exceptional generalization capability also enables it to be trained on small-scale graphs and directly infer the seed node set for large-scale graphs. IMGNN achieves these advantages by adopting a novel design for feature construction and model training, utilizing features constructed from influence propagations over graphs with randomly skipped nodes. This approach enables IMGNN to avoid overfitting to specific network structures while employing a unique technique to improve time efficiency by training on smaller networks. We have conducted extensive experiments using real-world social networks with up to 40 million nodes, and the results strongly demonstrate the superiority of IMGNN in terms of influence spread, seed node set inference speed, and generalizability.
- 20 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1904-1915
Open AccessIMGNN: An Efficient, Effective and Generalizable Algorithm for Influence Maximization in Social Networks
Influence Maximization (IM) is a crucial problem in social network analysis and has been extensively studied. Traditional approaches rely on designing approximation algorithms using network sampling; however, these methods lack generalizability and depend on an explicit definition of the influence diffusion model as input. Recently, researchers have turned to deep learning methods to address the shortcomings of traditional IM algorithms, but current learning-based IM algorithms still suffer from severe deficiencies in scalability and generalizability. In this paper, we propose IMGNN, a simple, efficient, effective, and generalizable algorithm powered by graph neural networks. IMGNN is a learning-based IM algorithm with strong generalization capability that reduces the overhead of model retraining while also providing fast seed set inference speed. As a result, IMGNN achieves better performance in terms of both efficiency and effectiveness compared to existing IM algorithms. Its exceptional generalization capability also enables it to be trained on small-scale graphs and directly infer the seed node set for large-scale graphs. IMGNN achieves these advantages by adopting a novel design for feature construction and model training, utilizing features constructed from influence propagations over graphs with randomly skipped nodes. This approach enables IMGNN to avoid overfitting to specific network structures while employing a unique technique to improve time efficiency by training on smaller networks. We have conducted extensive experiments using real-world social networks with up to 40 million nodes, and the results strongly demonstrate the superiority of IMGNN in terms of influence spread, seed node set inference speed, and generalizability.
- 8 May 2025
- Published by Elsevier in Artificial Intelligence
- Vol. 345
Efficient and effective budget-feasible mechanisms for submodular valuations
- 8 May 2025
- Published by Elsevier in Artificial Intelligence
- Vol. 345
Efficient and effective budget-feasible mechanisms for submodular valuations
- Open Access
- 22 April 2025
- Published by Association for Computing Machinery (ACM)
- p. 4710-4721
Linear-Time Algorithms for Representative Subset Selection From Data Streams
- 22 April 2025
- Published by Association for Computing Machinery (ACM)
- p. 4710-4721
Open AccessLinear-Time Algorithms for Representative Subset Selection From Data Streams
- 10 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. 12 (6) , 4517-4531
Modality-Disentangled Feature Extraction via Knowledge Distillation in Multimodal Recommendation Systems
Multimodal recommendation enhances item representation in recommendation systems by integrating diverse modalities of item information beyond traditional ID-based features. This approach utilizes supplementary item details, including images, text, videos, and audio, to refine the accuracy of item representation and thereby boost the precision of recommendations. Multimodal recommendation has emerged as a vibrant field within the scope of systems that generate suggestions. It offers a powerful approach to address the challenges of data scarcity and the representation of long-tail content, thereby improving the overall quality of recommendations. However, the advancement in multimodal recommendation is currently hindered by two main obstacles. First, the process of extracting multimodal features from pre-trained models using either shallow or deep neural networks often results in insufficient data extraction or sparse recommendation data, leading to suboptimal model performance. Second, a significant portion of previous research has focused on integrating information across modalities, often overlooking the distinct characteristics inherent in different modalities. Addressing these challenges, we introduce a unique methodology titled “modality-disentangled feature extraction via knowledge distillation in multimodal recommendation systems” (MODEST). First, to tackle the aforementioned problems that arise when extracting multimodal features with either deep or shallow neural networks, our approach adopts a teacher–student network framework. In this framework, deep neural networks are utilized to extract representation vectors from text and image data. Feature fusion is then carried out via attention mechanisms, and semantic labels are employed as classification labels to derive three supervised learning loss functions. This process significantly enhances the teacher network’s capacity to extract multimodal features. Subsequently, the knowledge from the teacher network is transferred to the student network through knowledge distillation. The student network makes use of a shallow neural network, and during the inference stage, we rely on the student network. This strategy effectively resolves the issues of data sparsity in deep networks and insufficient information extraction in shallow networks. Second, to more effectively capture the similarities and distinct features among different modalities, we implement a disentangled modality decomposition technique. Through integrated mappings, it separately extracts text and image information from the teacher–student networks and decomposes them into cross-modality common information and cross-modality specific information. By applying the constraints of contrastive learning, we minimize the distance between cross-modality common information and maximize the separation of cross-modality specific information, promoting convergence with the aid of auxiliary loss. This effectively addresses the problem of cross-modality feature alignment. Lastly, we combine the recommendation loss function with the multiple loss constraints we have added to formulate a unified optimization objective function. To underscore the remarkable efficacy of our proposed model, we have executed comprehensive experiments and visualizations on several real-world datasets. The results distinctly show a significant enhancement in our model’s performance, allowing it to achieve a level of competitiveness against other methods.
- 10 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. 12 (6) , 4517-4531
Modality-Disentangled Feature Extraction via Knowledge Distillation in Multimodal Recommendation Systems
Multimodal recommendation enhances item representation in recommendation systems by integrating diverse modalities of item information beyond traditional ID-based features. This approach utilizes supplementary item details, including images, text, videos, and audio, to refine the accuracy of item representation and thereby boost the precision of recommendations. Multimodal recommendation has emerged as a vibrant field within the scope of systems that generate suggestions. It offers a powerful approach to address the challenges of data scarcity and the representation of long-tail content, thereby improving the overall quality of recommendations. However, the advancement in multimodal recommendation is currently hindered by two main obstacles. First, the process of extracting multimodal features from pre-trained models using either shallow or deep neural networks often results in insufficient data extraction or sparse recommendation data, leading to suboptimal model performance. Second, a significant portion of previous research has focused on integrating information across modalities, often overlooking the distinct characteristics inherent in different modalities. Addressing these challenges, we introduce a unique methodology titled “modality-disentangled feature extraction via knowledge distillation in multimodal recommendation systems” (MODEST). First, to tackle the aforementioned problems that arise when extracting multimodal features with either deep or shallow neural networks, our approach adopts a teacher–student network framework. In this framework, deep neural networks are utilized to extract representation vectors from text and image data. Feature fusion is then carried out via attention mechanisms, and semantic labels are employed as classification labels to derive three supervised learning loss functions. This process significantly enhances the teacher network’s capacity to extract multimodal features. Subsequently, the knowledge from the teacher network is transferred to the student network through knowledge distillation. The student network makes use of a shallow neural network, and during the inference stage, we rely on the student network. This strategy effectively resolves the issues of data sparsity in deep networks and insufficient information extraction in shallow networks. Second, to more effectively capture the similarities and distinct features among different modalities, we implement a disentangled modality decomposition technique. Through integrated mappings, it separately extracts text and image information from the teacher–student networks and decomposes them into cross-modality common information and cross-modality specific information. By applying the constraints of contrastive learning, we minimize the distance between cross-modality common information and maximize the separation of cross-modality specific information, promoting convergence with the aid of auxiliary loss. This effectively addresses the problem of cross-modality feature alignment. Lastly, we combine the recommendation loss function with the multiple loss constraints we have added to formulate a unified optimization objective function. To underscore the remarkable efficacy of our proposed model, we have executed comprehensive experiments and visualizations on several real-world datasets. The results distinctly show a significant enhancement in our model’s performance, allowing it to achieve a level of competitiveness against other methods.
- Open Access
- 24 August 2024
- Published by Association for Computing Machinery (ACM)
- p. 514-525
Fairness in Streaming Submodular Maximization Subject to a Knapsack Constraint
Submodular optimization has been identified as a powerful tool for many data mining applications, where a representative subset of moderate size needs to be extracted from a large-scale dataset. In scenarios where data points possess sensitive attributes such as age, gender, or race, it becomes imperative to integrate fairness measures into submodular optimization to mitigate bias and discrimination. In this paper, we study the fundamental problem of fair submodular maximization subject to a knapsack constraint and propose the first streaming algorithm for it with provable performance guarantees for both monotone and non-monotone submodular functions. As a byproduct, we also propose a streaming algorithm for submodular maximization subject to a partition matroid and a knapsack constraint, significantly improving the performance bounds achieved by previous work. We conduct extensive experiments on real-world applications such as movie recommendation, image summarization, and maximum coverage in social networks. The experimental results strongly demonstrate the superiority of our proposed algorithms in terms of both fairness and utility.
- 24 August 2024
- Published by Association for Computing Machinery (ACM)
- p. 514-525
Open AccessFairness in Streaming Submodular Maximization Subject to a Knapsack Constraint
Submodular optimization has been identified as a powerful tool for many data mining applications, where a representative subset of moderate size needs to be extracted from a large-scale dataset. In scenarios where data points possess sensitive attributes such as age, gender, or race, it becomes imperative to integrate fairness measures into submodular optimization to mitigate bias and discrimination. In this paper, we study the fundamental problem of fair submodular maximization subject to a knapsack constraint and propose the first streaming algorithm for it with provable performance guarantees for both monotone and non-monotone submodular functions. As a byproduct, we also propose a streaming algorithm for submodular maximization subject to a partition matroid and a knapsack constraint, significantly improving the performance bounds achieved by previous work. We conduct extensive experiments on real-world applications such as movie recommendation, image summarization, and maximum coverage in social networks. The experimental results strongly demonstrate the superiority of our proposed algorithms in terms of both fairness and utility.
- 23 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. 11 (5) , 6087-6097
Spatial and Temporal User Interest Representations for Sequential Recommendation
In recent years, recommendation systems have become increasingly prevalent in various fields, facilitating quick access to the information users need. As a result, many models have been proposed to model user interests, leading to more accurate recommendation lists, superior user experience, and business value. However, characterizing the dynamically changing interests of users is a challenging task. User interests shift over time while maintaining some long-term interests, and at each time, users’ interests are diverse. To investigate the benefits of multidimensional interests for users, this article proposes to characterize user preferences based on their spatiotemporal interests. Utilizing temporal and spatial information is critical for improving recommendation accuracy. To achieve this, we present a novel approach called multilong short-term interest (MLSI) user representation for recommendation. This method extracts long-term and short-term interests of users from their behavioural sequences using decoupled self-supervised learning with different optimizers. Self-attention is then employed to capture the diverse interests of users through their behavioral sequences. Final, long-term and short-term interests, as well as diversified interests, are aggregated to represent user interests. Extensive experiments on real-world datasets show that MLSI not only outperforms state-of-the-art methods but also more effectively characterizes user interests, reflecting an improvement ranging from 5% to 20% across various metrics on multiple datasets.
- 23 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. 11 (5) , 6087-6097
Spatial and Temporal User Interest Representations for Sequential Recommendation
In recent years, recommendation systems have become increasingly prevalent in various fields, facilitating quick access to the information users need. As a result, many models have been proposed to model user interests, leading to more accurate recommendation lists, superior user experience, and business value. However, characterizing the dynamically changing interests of users is a challenging task. User interests shift over time while maintaining some long-term interests, and at each time, users’ interests are diverse. To investigate the benefits of multidimensional interests for users, this article proposes to characterize user preferences based on their spatiotemporal interests. Utilizing temporal and spatial information is critical for improving recommendation accuracy. To achieve this, we present a novel approach called multilong short-term interest (MLSI) user representation for recommendation. This method extracts long-term and short-term interests of users from their behavioural sequences using decoupled self-supervised learning with different optimizers. Self-attention is then employed to capture the diverse interests of users through their behavioral sequences. Final, long-term and short-term interests, as well as diversified interests, are aggregated to represent user interests. Extensive experiments on real-world datasets show that MLSI not only outperforms state-of-the-art methods but also more effectively characterizes user interests, reflecting an improvement ranging from 5% to 20% across various metrics on multiple datasets.
- Open Access
- 30 December 2023
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Information Systems
- Vol. 42 (3) , 1-30
H3GNN: Hybrid Hierarchical HyperGraph Neural Network for Personalized Session-based Recommendation
Personalized Session-based recommendation (PSBR) is a general and challenging task in the real world, aiming to recommend a session’s next clicked item based on the session’s item transition information and the corresponding user’s historical sessions. A session is defined as a sequence of interacted items during a short period. The PSBR problem has a natural hierarchical architecture in which each session consists of a series of items, and each user owns a series of sessions. However, the existing PSBR methods can merely capture the pairwise relation information within items and users. To effectively capture the hierarchical information, we propose a novel hierarchical hypergraph neural network to model the hierarchical architecture. Moreover, considering that the items in sessions are sequentially ordered, while the hypergraph can only model the set relation, we propose a directed graph aggregator (DGA) to aggregate the sequential information from the directed global item graph. By attentively combining the embeddings of the above two modules, we propose a framework dubbed H3GNN (Hybrid Hierarchical HyperGraph Neural Network). Extensive experiments on three benchmark datasets demonstrate the superiority of our proposed model compared to the state-of-the-art methods, and ablation experiment results validate the effectiveness of all the proposed components.
- 30 December 2023
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Information Systems
- Vol. 42 (3) , 1-30
Open AccessH3GNN: Hybrid Hierarchical HyperGraph Neural Network for Personalized Session-based Recommendation
Personalized Session-based recommendation (PSBR) is a general and challenging task in the real world, aiming to recommend a session’s next clicked item based on the session’s item transition information and the corresponding user’s historical sessions. A session is defined as a sequence of interacted items during a short period. The PSBR problem has a natural hierarchical architecture in which each session consists of a series of items, and each user owns a series of sessions. However, the existing PSBR methods can merely capture the pairwise relation information within items and users. To effectively capture the hierarchical information, we propose a novel hierarchical hypergraph neural network to model the hierarchical architecture. Moreover, considering that the items in sessions are sequentially ordered, while the hypergraph can only model the set relation, we propose a directed graph aggregator (DGA) to aggregate the sequential information from the directed global item graph. By attentively combining the embeddings of the above two modules, we propose a framework dubbed H3GNN (Hybrid Hierarchical HyperGraph Neural Network). Extensive experiments on three benchmark datasets demonstrate the superiority of our proposed model compared to the state-of-the-art methods, and ablation experiment results validate the effectiveness of all the proposed components.