Hong Peng

Dr. Hong Peng

software engineering, Yunnan University, Kunming, Yunnan, China
orcid

Research Focus & Metrics

Academic Career Length

20Years (2007-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 February 2020 in Neurocomputing
Verified
Journal Article
Closed access
WWWanlin Weng
Wanlin Weng
WZWeiwei Zhou
Weiwei Zhou
JCJiazhou Chen
Jiazhou Chen
Hong PengHong Peng
Hong Peng
HCHongmin Cai
Hongmin Cai

24 November 2020 in IEEE Transactions on Pattern Analysis and Machine Intelligence
Verified
Journal Article
Closed access
Hong PengHong Peng
Hong Peng
YHYu Hu
Yu Hu
JCJiazhou Chen
Jiazhou Chen
HWHaiyan Wang
Haiyan Wang
YLYang Li
Yang Li
HCHongmin Cai
Hongmin Cai
The performance of most the clustering methods hinges on the used pairwise affinity, which is usually denoted by a similarity matrix. However, the pairwise similarity is notoriously known for its venerability of noise contamination or the imbalance in samples or features, and thus hinders accurate clustering. To tackle this issue, we propose to use information among samples to boost the clustering performance. We proved that a simplified similarity for pairs, denoted by a fourth order tensor, equals to the Kronecker product of pairwise similarity matrices under decomposable assumption, or provide complementary information for which the pairwise similarity missed under indecomposable assumption. Then a high order similarity matrix is obtained from the tensor similarity via eigenvalue decomposition. The high order similarity capturing spatial information serves as a robust complement for the pairwise similarity. It is further integrated with the popular pairwise similarity, named by IPS2, to boost the clustering performance. Extensive experiments demonstrated that the proposed IPS2 significantly outperformed previous similarity-based methods on real-world datasets and it was capable of handling the clustering task over under-sampled and noisy datasets.

1 December 2007
Verified
Conference Paper
Closed access
YLYishen Lin
Yishen Lin
Hong PengHong Peng
Hong Peng
A hybrid niching gene expression programming algorithm, which combines the niching method and clustering model, is proposed. Similar to other evolution algorithms, GEP also has the problem of premature convergence. Niching method is critical to keep diversity among the population and to use this diversity as resource for exploratory evolution. K-means clustering algorithm was used to cluster the near individuals and to build the niche. This model can make GEP jump out of the local optimization at a greater probability and find the global optimization. Experimental results on function finding problems show that the algorithm has higher precision and better search ability than the basic GEP.

7 January 2026 in Knowledge and Information Systems
Verified
Journal Article
Closed access
PDF file
YLYuanxian Li
Yuanxian Li
YDYuan Du
Yuan Du
Hong PengHong Peng
Hong Peng
ZHZhenli He
Zhenli He
XJXin Jin
Xin Jin
CXCheng Xie
Cheng Xie

17 June 2025 in IEEE Transactions on Computational Social Systems
Verified
Journal Article
Closed access
Hong PengHong Peng
Hong Peng
CFChanglong Fu
Changlong Fu
XZXiong Zhang
Xiong Zhang
CXCheng Xie
Cheng Xie
HCHongming Cai
Hongming Cai
Weiming ShenWeiming Shen
Weiming Shen
Multimodal recommendation has become a key technology for social media platforms. It is widely used in content recommendation, user preference analysis, advertisement placement, etc. Existing recommendation methods mainly focus on learning multimodal embeddings from direct interactions between users and items, ignoring indirect interactions among users-to-users and items-to-items. This limits the further exploration of potential interests between users and items. Moreover, during the model training, classical recommendation methods usually randomly select uninteracted items of a user as their negative samples. This may introduce significant learning bias, as uninteracted items could be false negatives and still potentially interest the user. To this end, we propose a novel indirect interactions discovery and true negative sampling multimodal recommendation (ITMRec) method to further explore potential user interests and mitigate the issue of false negative samples during learning. Specifically, we propose an indirect interactions discovering (IID) model to explore the latent interests among users-to-users and items-to-items. Then, we propose a true negative sampling (TNS) model to refine negative sampling that can alleviate the false negative sample problem. Finally, we enhance existing collaborative filtering methods by integrating representations derived from multimodal content, indirect interactions discovery, and refined negative sampling strategies, allowing for more precise alignment with users’ latent interests. Extensive experiments on three benchmark datasets demonstrate that our ITMRec significantly outperforms state-of-the-art recommendation baselines, achieving a 3.64% improvement over peer methods. The code is available at https://github.com/long-best/ITMRec.git.

2 February 2026 in IEEE Transactions on Computational Social Systems
Verified
Journal Article
Closed access
ZWZiyan Wang
Ziyan Wang
Hong PengHong Peng
Hong Peng
CFChanglong Fu
Changlong Fu
XZXiong Zhang
Xiong Zhang
CXCheng Xie
Cheng Xie
YYYun Yang
Yun Yang
Multimodal recommendation is a crucial technology on social media platforms. It is widely applied in scenarios such as product recommendation and advertising delivery. However, existing multimodal recommendation approaches often overlook invariant semantic features that persist across modalities, leading to decreased robustness and generalization. To address this limitation, we propose GILMRec, a novel graph invariant learning-based multimodal social media recommendation framework. The GILMRec introduces an invariant feature learning strategy to extract invariant features separately from visual and textual modalities and employs an attention-based fusion mechanism to integrate them into a unified embedding. Specifically, we construct modality-specific similarity graphs and apply top- $t$ neighbor aggregation, enhancing the consistency of invariant features while effectively suppressing modality-specific noise. Extensive experiments on three Amazon benchmark datasets and a large-scale dataset [baby, sports, clothing, and compact discs (CDs)] demonstrate that GILMRec consistently outperforms twelve state-of-the-art baselines. The results confirm the efficiency of invariant features in capturing robust multimodal representations and improving recommendation performance, particularly in sparse data scenarios.