Hongji Shu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Hongji Shu has a total of 8 co-authors. Here are the top co-authors by number of shared publications:
- Jia Zhu
Zhejiang Key Laboratory of Intelligent Education Technology and Application, Zhejiang Normal University, Jinhua, China
2 shared publications Chaojun Meng
School of Computer Science and Technology, Zhejiang Normal University, Jinhua, China
1 shared publication- Cong Zhou
School of Computer Science and Technology, Zhejiang Normal University, Jinhua, China
1 shared publication - Jianyang Shi
Zhejiang Normal University
1 shared publication - Kwame Omono Asamoah
School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, China
1 shared publication - Pasquale De Meo
University of Messina
1 shared publication
Latest Publications
- Open Access
- 2 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 20830-20839
Self-Supervised Hypergraph Learning for Enhanced Multimodal Representation
Hypergraph neural networks have gained substantial popularity in capturing complex correlations between data items in multimodal datasets. In this study, we propose a novel approach called the self-supervised hypergraph learning (SHL) framework that focuses on extracting hypergraph features to improve multimodal representation. Our method utilizes a dual embedding strategy and leverages SHL to improve the accuracy and robustness of the model. To achieve this, we employ a hypergraph learning framework to extract global context effectively by capturing rich inter-modal dependencies. Additionally, we introduce a novel self-supervised learning (SSL) component that utilizes the interaction graph data, thereby strengthening the robustness of the model. By jointly optimizing hypergraph feature extraction and SSL, SHL significantly improves the performance of multimodal representation tasks. To validate the effectiveness of our approach, we construct two comprehensive multimodal micro-video recommendation datasets using publicly available data (TikTok and MovieLens-10M). Prior to dataset creation, we meticulously handle invalid entries and outliers and complete missing mode information using external auxiliary sources, such as YouTube. These datasets are made publicly available to the research community for evaluation purposes. Experimental results on the above recommendation datasets demonstrate that the proposed SHL approach outperforms state-of-the-art baselines, highlighting its superior performance in multimodal representation tasks.
- 2 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 20830-20839
Open AccessSelf-Supervised Hypergraph Learning for Enhanced Multimodal Representation
Hypergraph neural networks have gained substantial popularity in capturing complex correlations between data items in multimodal datasets. In this study, we propose a novel approach called the self-supervised hypergraph learning (SHL) framework that focuses on extracting hypergraph features to improve multimodal representation. Our method utilizes a dual embedding strategy and leverages SHL to improve the accuracy and robustness of the model. To achieve this, we employ a hypergraph learning framework to extract global context effectively by capturing rich inter-modal dependencies. Additionally, we introduce a novel self-supervised learning (SSL) component that utilizes the interaction graph data, thereby strengthening the robustness of the model. By jointly optimizing hypergraph feature extraction and SSL, SHL significantly improves the performance of multimodal representation tasks. To validate the effectiveness of our approach, we construct two comprehensive multimodal micro-video recommendation datasets using publicly available data (TikTok and MovieLens-10M). Prior to dataset creation, we meticulously handle invalid entries and outliers and complete missing mode information using external auxiliary sources, such as YouTube. These datasets are made publicly available to the research community for evaluation purposes. Experimental results on the above recommendation datasets demonstrate that the proposed SHL approach outperforms state-of-the-art baselines, highlighting its superior performance in multimodal representation tasks.
- Open Access
GUDN: A novel guide network with label reinforcement strategy for extreme multi-label text classification
In natural language processing, extreme multi-label text classification is an emerging but essential task. The problem of extreme multi-label text classification (XMTC) is to recall some of the most relevant labels for a text from an extremely large label set. Large-scale pre-trained models have brought a new trend to this problem. Though the large-scale pre-trained models have made significant achievements on this problem, the valuable fine-tuned methods have yet to be studied. Though label semantics have been introduced in XMTC, the vast semantic gap between texts and labels has yet to gain enough attention. This paper builds a new guide network (GUDN) to help fine-tune the pre-trained model to instruct classification later. Furthermore, GUDN uses raw label semantics combined with a helpful label reinforcement strategy to effectively explore the latent space between texts and labels, narrowing the semantic gap, which can further improve predicted accuracy. Experimental results demonstrate that GUDN outperforms state-of-the-art methods on Eurlex-4k and has competitive results on other popular datasets. In an additional experiment, we investigated the input lengths' influence on the Transformer-based model's accuracy. Our source code is released at https://t.hk.uy/aFSH.
Open AccessGUDN: A novel guide network with label reinforcement strategy for extreme multi-label text classification
In natural language processing, extreme multi-label text classification is an emerging but essential task. The problem of extreme multi-label text classification (XMTC) is to recall some of the most relevant labels for a text from an extremely large label set. Large-scale pre-trained models have brought a new trend to this problem. Though the large-scale pre-trained models have made significant achievements on this problem, the valuable fine-tuned methods have yet to be studied. Though label semantics have been introduced in XMTC, the vast semantic gap between texts and labels has yet to gain enough attention. This paper builds a new guide network (GUDN) to help fine-tune the pre-trained model to instruct classification later. Furthermore, GUDN uses raw label semantics combined with a helpful label reinforcement strategy to effectively explore the latent space between texts and labels, narrowing the semantic gap, which can further improve predicted accuracy. Experimental results demonstrate that GUDN outperforms state-of-the-art methods on Eurlex-4k and has competitive results on other popular datasets. In an additional experiment, we investigated the input lengths' influence on the Transformer-based model's accuracy. Our source code is released at https://t.hk.uy/aFSH.