Kai Sun
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 Sun has a total of 16 co-authors. Here are the top co-authors by number of shared publications:
- Samuel Mensah
University of Sheffield
4 shared publications - Richong Zhang
Beihang University, Beijing, China
3 shared publications - Yongyi Mao
University of Ottawa UNIVERSITY OF OTTAWA UNIVERSITY OF OTTAWA
3 shared publications - Bin Shi
Xi'an Jiaotong University, China
2 shared publications - Bo Dong
School of Distance Education, Xi'an Jiaotong University, Xi'an, China
2 shared publications - Xudong Liu
Beihang University, Beijing, China
2 shared publications
Latest Publications
- 12 December 2025
- Published by Elsevier in Information Processing & Management
- Vol. 63 (3)
RADAR: Relation-assisted dual-graph aligning recognition for grounded multimodal named entity recognition
- 12 December 2025
- Published by Elsevier in Information Processing & Management
- Vol. 63 (3)
RADAR: Relation-assisted dual-graph aligning recognition for grounded multimodal named entity recognition
- 28 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Audio, Speech and Language Processing
- Vol. 33, 1324-1336
Dual-Aspect Noise-Based Regularization for Multi-Modal Relation Extraction in Media Posts
Multi-Modal Relation Extraction (MMRE) plays a key role in various multimedia applications including, recommendation and information retrieval systems. MMRE aims to extract the semantic relation between entities by leveraging context from a text-image pair. By utilizing context from images, the challenge of learning from noisy images in MMRE emerges as a research problem by itself. For instance, subtle variations in similar images can act as noise and potentially impact the predictions made by MMRE models. To tackle this problem, current work utilizes attention mechanisms to fuse relevant text and image features or devise data augmentation techniques (e.g., via generative models) to improve generalization. However, the current performance still remains unsatisfactory. In an effort to improve upon the performance, we propose a Dual-Aspect Noise-based Regularization framework that encompasses two techniques: 1) noise removal through an adaptive gating mechanism, 2) fighting noise with noise to improve feature stability in the learning process. We find that combining these techniques encourages the model to focus on more relevant image features for MMRE. We carry out extensive experiments and demonstrate that our proposed model is further enhanced by exploring data augmentation techniques. This additional improvement leads the model to achieve state-of-the-art performance on the widely-used Multi-modal Neural Relation Extraction (MNRE) dataset, and show its effectiveness and generalizability on the Multi-Modal Named Entity Recognition task.
- 28 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Audio, Speech and Language Processing
- Vol. 33, 1324-1336
Dual-Aspect Noise-Based Regularization for Multi-Modal Relation Extraction in Media Posts
Multi-Modal Relation Extraction (MMRE) plays a key role in various multimedia applications including, recommendation and information retrieval systems. MMRE aims to extract the semantic relation between entities by leveraging context from a text-image pair. By utilizing context from images, the challenge of learning from noisy images in MMRE emerges as a research problem by itself. For instance, subtle variations in similar images can act as noise and potentially impact the predictions made by MMRE models. To tackle this problem, current work utilizes attention mechanisms to fuse relevant text and image features or devise data augmentation techniques (e.g., via generative models) to improve generalization. However, the current performance still remains unsatisfactory. In an effort to improve upon the performance, we propose a Dual-Aspect Noise-based Regularization framework that encompasses two techniques: 1) noise removal through an adaptive gating mechanism, 2) fighting noise with noise to improve feature stability in the learning process. We find that combining these techniques encourages the model to focus on more relevant image features for MMRE. We carry out extensive experiments and demonstrate that our proposed model is further enhanced by exploring data augmentation techniques. This additional improvement leads the model to achieve state-of-the-art performance on the widely-used Multi-modal Neural Relation Extraction (MNRE) dataset, and show its effectiveness and generalizability on the Multi-Modal Named Entity Recognition task.
- Open Access
- 13 May 2024
- Published by Association for Computing Machinery (ACM)
- p. 4283-4294
Multimodal Relation Extraction via a Mixture of Hierarchical Visual Context Learners
Multimodal relation extraction is a fundamental task of multimodal information extraction. Recent studies have shown promising results by integrating hierarchical visual features from local regions, like image patches, to the broader global regions that form the entire image. However, research to date has largely ignored the understanding of how hierarchical visual semantics are represented and the characteristics that can benefit relation extraction. To bridge this gap, we propose a novel two-stage hierarchical visual context fusion transformer incorporating the mixture of multimodal experts framework to effectively represent and integrate hierarchical visual features into textual semantic representations. In addition, we introduce the concept of hierarchical tracking maps to facilitate the understanding of the intrinsic mechanisms of image information processing involved in multimodal models. We thoroughly investigate the implications of hierarchical visual contexts through four dimensions: performance evaluation, the nature of auxiliary visual information, the patterns observed in the image encoding hierarchy, and the significance of various visual encoding levels. Empirical studies show that our approach achieves new state-of-the-art performance on the MNRE dataset.
- 13 May 2024
- Published by Association for Computing Machinery (ACM)
- p. 4283-4294
Open AccessMultimodal Relation Extraction via a Mixture of Hierarchical Visual Context Learners
Multimodal relation extraction is a fundamental task of multimodal information extraction. Recent studies have shown promising results by integrating hierarchical visual features from local regions, like image patches, to the broader global regions that form the entire image. However, research to date has largely ignored the understanding of how hierarchical visual semantics are represented and the characteristics that can benefit relation extraction. To bridge this gap, we propose a novel two-stage hierarchical visual context fusion transformer incorporating the mixture of multimodal experts framework to effectively represent and integrate hierarchical visual features into textual semantic representations. In addition, we introduce the concept of hierarchical tracking maps to facilitate the understanding of the intrinsic mechanisms of image information processing involved in multimodal models. We thoroughly investigate the implications of hierarchical visual contexts through four dimensions: performance evaluation, the nature of auxiliary visual information, the patterns observed in the image encoding hierarchy, and the significance of various visual encoding levels. Empirical studies show that our approach achieves new state-of-the-art performance on the MNRE dataset.
- Open Access
- 30 April 2023
- Published by Association for Computing Machinery (ACM)
- p. 1594-1603
Self-training through Classifier Disagreement for Cross-Domain Opinion Target Extraction
Opinion target extraction (OTE) or aspect extraction (AE) is a fundamental task in opinion mining that aims to extract the targets (or aspects) on which opinions have been expressed. Recent work focus on cross-domain OTE, which is typically encountered in real-world scenarios, where the testing and training distributions differ. Most methods use domain adversarial neural networks that aim to reduce the domain gap between the labelled source and unlabelled target domains to improve target domain performance. However, this approach only aligns feature distributions and does not account for class-wise feature alignment, leading to suboptimal results. Semi-supervised learning (SSL) has been explored as a solution, but is limited by the quality of pseudo-labels generated by the model. Inspired by the theoretical foundations in domain adaptation [2], we propose a new SSL approach that opts for selecting target samples whose model output from a domain-specific teacher and student network disagree on the unlabelled target data, in an effort to boost the target domain performance. Extensive experiments on benchmark cross-domain OTE datasets show that this approach is effective and performs consistently well in settings with large domain shifts.
- 30 April 2023
- Published by Association for Computing Machinery (ACM)
- p. 1594-1603
Open AccessSelf-training through Classifier Disagreement for Cross-Domain Opinion Target Extraction
Opinion target extraction (OTE) or aspect extraction (AE) is a fundamental task in opinion mining that aims to extract the targets (or aspects) on which opinions have been expressed. Recent work focus on cross-domain OTE, which is typically encountered in real-world scenarios, where the testing and training distributions differ. Most methods use domain adversarial neural networks that aim to reduce the domain gap between the labelled source and unlabelled target domains to improve target domain performance. However, this approach only aligns feature distributions and does not account for class-wise feature alignment, leading to suboptimal results. Semi-supervised learning (SSL) has been explored as a solution, but is limited by the quality of pseudo-labels generated by the model. Inspired by the theoretical foundations in domain adaptation [2], we propose a new SSL approach that opts for selecting target samples whose model output from a domain-specific teacher and student network disagree on the unlabelled target data, in an effort to boost the target domain performance. Extensive experiments on benchmark cross-domain OTE datasets show that this approach is effective and performs consistently well in settings with large domain shifts.
- Open Access
- 8 April 2023
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Information Systems
- Vol. 41 (2) , 1-29
Learning Implicit and Explicit Multi-task Interactions for Information Extraction
Information extraction aims at extracting entities, relations, and so on, in text to support information retrieval systems. To extract information, researchers have considered multitask learning (ML) approaches. The conventional ML approach learns shared features across tasks, with the assumption that these features capture sufficient task interactions to learn expressive shared representations for task classification. However, such an assumption is flawed in different perspectives. First, the shared representation may contain noise introduced by another task; tasks coupled for multitask learning may have different complexities but this approach treats all tasks equally; the conventional approach has a flat structure that hinders the learning of explicit interactions. This approach, however, learns implicit interactions across tasks and often has a generalization ability that has benefited the learning of multitasks. In this article, we take advantage of implicit interactions learned by conventional approaches while alleviating the issues mentioned above by developing a Recurrent Interaction Network with an effective Early Prediction Integration (RIN-EPI) for multitask learning. Specifically, RIN-EPI learns implicit and explicit interactions across two different but related tasks. To effectively learn explicit interactions across tasks, we consider the correlations among the outputs of related tasks. It is, however, obvious that task outputs are unobservable during training, so we leverage the predictions at intermediate layers (referred to as early predictions) as proxies as well as shared features across tasks to learn explicit interactions through attention mechanisms and sequence learning models. By recurrently learning explicit interactions, we gradually improve predictions for the individual tasks in the multitask learning. We demonstrate the effectiveness of RIN-EPI on the learning of two mainstream multitasks for information extraction: (1) entity recognition and relation classification and (2) aspect and opinion term co-extraction. Extensive experiments demonstrate the effectiveness of the RIN-EPI architecture, where we achieve state-of-the-art results on several benchmark datasets.
- 8 April 2023
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Information Systems
- Vol. 41 (2) , 1-29
Open AccessLearning Implicit and Explicit Multi-task Interactions for Information Extraction
Information extraction aims at extracting entities, relations, and so on, in text to support information retrieval systems. To extract information, researchers have considered multitask learning (ML) approaches. The conventional ML approach learns shared features across tasks, with the assumption that these features capture sufficient task interactions to learn expressive shared representations for task classification. However, such an assumption is flawed in different perspectives. First, the shared representation may contain noise introduced by another task; tasks coupled for multitask learning may have different complexities but this approach treats all tasks equally; the conventional approach has a flat structure that hinders the learning of explicit interactions. This approach, however, learns implicit interactions across tasks and often has a generalization ability that has benefited the learning of multitasks. In this article, we take advantage of implicit interactions learned by conventional approaches while alleviating the issues mentioned above by developing a Recurrent Interaction Network with an effective Early Prediction Integration (RIN-EPI) for multitask learning. Specifically, RIN-EPI learns implicit and explicit interactions across two different but related tasks. To effectively learn explicit interactions across tasks, we consider the correlations among the outputs of related tasks. It is, however, obvious that task outputs are unobservable during training, so we leverage the predictions at intermediate layers (referred to as early predictions) as proxies as well as shared features across tasks to learn explicit interactions through attention mechanisms and sequence learning models. By recurrently learning explicit interactions, we gradually improve predictions for the individual tasks in the multitask learning. We demonstrate the effectiveness of RIN-EPI on the learning of two mainstream multitasks for information extraction: (1) entity recognition and relation classification and (2) aspect and opinion term co-extraction. Extensive experiments demonstrate the effectiveness of the RIN-EPI architecture, where we achieve state-of-the-art results on several benchmark datasets.
- Open Access
- 1 January 2015
- Published by Association for Computational Linguistics (ACL)
- p. 295-304
Recurrent Polynomial Network for Dialogue State Tracking with Mismatched Semantic Parsers
- 1 January 2015
- Published by Association for Computational Linguistics (ACL)
- p. 295-304
Open AccessRecurrent Polynomial Network for Dialogue State Tracking with Mismatched Semantic Parsers