Cangzhi Zheng
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Cangzhi Zheng has a total of 16 co-authors. Here are the top co-authors by number of shared publications:
- Junjie Peng
Shanghai University
4 shared publications Lan Wang
School of Computer Engineering and Science Shanghai University
3 shared publicationsYijie Jin
School of Computer Engineering and Science Shanghai University
2 shared publications- Fen Yi
National Engineering Laboratory for Logistics Information Technology, YTO Express Co., Ltd., Shanghai, China
1 shared publication - Feng Cheng
University of Potsdam
1 shared publication Haochen Yuan
School of Computer Engineering and Science Shanghai University
1 shared publication
Latest Publications
- Open Access
Multimodal Transformers are Hierarchical Modal-wise Heterogeneous Graphs
Multimodal Sentiment Analysis (MSA) is a rapidly developing field that integrates multimodal information to recognize sentiments, and existing models have made significant progress in this area. The central challenge in MSA is multimodal fusion, which is predominantly addressed by Multimodal Transformers (MulTs). Although act as the paradigm, MulTs suffer from efficiency concerns. In this work, from the perspective of efficiency optimization, we propose and prove that MulTs are hierarchical modal-wise heterogeneous graphs (HMHGs), and we introduce the graph-structured representation pattern of MulTs. Based on this pattern, we propose an Interlaced Mask (IM) mechanism to design the Graph-Structured and Interlaced-Masked Multimodal Transformer (GsiT). It is formally equivalent to MulTs which achieves an efficient weight-sharing mechanism without information disorder through IM, enabling All-Modal-In-One fusion with only 1/3 of the parameters of pure MulTs. A Triton kernel called Decomposition is implemented to ensure avoiding additional computational overhead. Moreover, it achieves significantly higher performance than traditional MulTs. To further validate the effectiveness of GsiT itself and the HMHG concept, we integrate them into multiple state-of-the-art models and demonstrate notable performance improvements and parameter reduction on widely used MSA datasets.
Open AccessMultimodal Transformers are Hierarchical Modal-wise Heterogeneous Graphs
Multimodal Sentiment Analysis (MSA) is a rapidly developing field that integrates multimodal information to recognize sentiments, and existing models have made significant progress in this area. The central challenge in MSA is multimodal fusion, which is predominantly addressed by Multimodal Transformers (MulTs). Although act as the paradigm, MulTs suffer from efficiency concerns. In this work, from the perspective of efficiency optimization, we propose and prove that MulTs are hierarchical modal-wise heterogeneous graphs (HMHGs), and we introduce the graph-structured representation pattern of MulTs. Based on this pattern, we propose an Interlaced Mask (IM) mechanism to design the Graph-Structured and Interlaced-Masked Multimodal Transformer (GsiT). It is formally equivalent to MulTs which achieves an efficient weight-sharing mechanism without information disorder through IM, enabling All-Modal-In-One fusion with only 1/3 of the parameters of pure MulTs. A Triton kernel called Decomposition is implemented to ensure avoiding additional computational overhead. Moreover, it achieves significantly higher performance than traditional MulTs. To further validate the effectiveness of GsiT itself and the HMHG concept, we integrate them into multiple state-of-the-art models and demonstrate notable performance improvements and parameter reduction on widely used MSA datasets.
- Open Access
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 2188-2209
Multimodal Transformers are Hierarchical Modal-wise Heterogeneous Graphs
Yijie Jin, Junjie Peng, Xuanchao Lin, Haochen Yuan, Lan Wang, Cangzhi Zheng. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 2188-2209
Open AccessMultimodal Transformers are Hierarchical Modal-wise Heterogeneous Graphs
Yijie Jin, Junjie Peng, Xuanchao Lin, Haochen Yuan, Lan Wang, Cangzhi Zheng. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.
- 23 August 2024
- Published by Elsevier in Expert Systems with Applications
- Vol. 258
Frame-level nonverbal feature enhancement based sentiment analysis
- 23 August 2024
- Published by Elsevier in Expert Systems with Applications
- Vol. 258
Frame-level nonverbal feature enhancement based sentiment analysis
- 14 August 2024
- Published by Elsevier in Knowledge-Based Systems
- Vol. 302
Extracting method for fine-grained emotional features in videos
- 14 August 2024
- Published by Elsevier in Knowledge-Based Systems
- Vol. 302
Extracting method for fine-grained emotional features in videos
- Open Access
- 25 March 2023
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Asian and Low-Resource Language Information Processing
- Vol. 22 (4) , 1-19
Short Text Classification of Chinese with Label Information Assisting
As a common language form in oral communication, short text is hard to be used in the applications such as intent understanding, text classification and so on due to its limited content and information, as well as irregular expression and missing components. To increase the availability of short texts in real applications, we propose a Label Information Assisting-based Model (LIAM) for Chinese short text classification. In the model, we jointly use sentence-level features and word-level features to reduce text information loss. And the sentence-level features are fused with relevant label information by the Label Information Extending and Fusion (LIEF) module while the word-level features are also enhanced with assistance of relevant label information. By utilizing the text-related information from labels as extended information, the model enriches and enhances the features of short text, benefiting classification. To verify the correctness and effectiveness of the proposed method, we conduct extensive experiments on four Chinese datasets and six sub-datasets with different models. The experimental results show that LIAM presented can effectively enrich information for text and much improve the performance of short text classification. It performs much better than other methods do. What is more, the less the training set, the greater the advantages of the model.
- 25 March 2023
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Asian and Low-Resource Language Information Processing
- Vol. 22 (4) , 1-19
Open AccessShort Text Classification of Chinese with Label Information Assisting
As a common language form in oral communication, short text is hard to be used in the applications such as intent understanding, text classification and so on due to its limited content and information, as well as irregular expression and missing components. To increase the availability of short texts in real applications, we propose a Label Information Assisting-based Model (LIAM) for Chinese short text classification. In the model, we jointly use sentence-level features and word-level features to reduce text information loss. And the sentence-level features are fused with relevant label information by the Label Information Extending and Fusion (LIEF) module while the word-level features are also enhanced with assistance of relevant label information. By utilizing the text-related information from labels as extended information, the model enriches and enhances the features of short text, benefiting classification. To verify the correctness and effectiveness of the proposed method, we conduct extensive experiments on four Chinese datasets and six sub-datasets with different models. The experimental results show that LIAM presented can effectively enrich information for text and much improve the performance of short text classification. It performs much better than other methods do. What is more, the less the training set, the greater the advantages of the model.