Journal Articles

1. Yulin Du , Haichuan Fang , Xin Li , Hongfu Zuo ^* , Xincan Zhao ^*

   SPPL: Predicting remaining useful life of aircraft engine via structural prior-guided patch-wise learning

   Expert Systems with Applications , pp. 132864 (2026)

   DOI BIB Abstract

   @article{du2026sppl,
     title = {SPPL: Predicting remaining useful life of aircraft engine via structural prior-guided patch-wise learning},
     author = {Du, Yulin and Fang, Haichuan and Li, Xin and Zuo, Hongfu and Zhao, Xincan},
     journal = {Expert Systems with Applications},
     pages = {132864},
     doi = {10.1016/j.eswa.2026.132864},
     correspondingauthor = {Hongfu Zuo|Xincan Zhao},
     year = {2026}
   }
   

   This work introduces structural prior-guided patch-wise learning for aircraft-engine remaining useful life prediction. The method divides long sensor sequences into patches, models local degradation through an intra-patch spatial-temporal graph guided by engine topology, and captures long-range evolution through an inter-patch relational graph. Fusing sensor-level and patch-level representations improves prediction accuracy and robustness on two benchmark datasets.

2. Haichuan Fang , Zhen Tian , Youwei Wang , Bo Ji , Yangdong Ye ^*

   Contrastive Variational Graph Symmetric Autoencoder for Full Extrapolation Over Temporal Knowledge Graphs

   IEEE Transactions on Systems, Man, and Cybernetics: Systems (2026)

   DOI BIB Abstract

   @article{fang2026contrastive,
     title = {Contrastive Variational Graph Symmetric Autoencoder for Full Extrapolation Over Temporal Knowledge Graphs},
     author = {Fang, Haichuan and Tian, Zhen and Wang, Youwei and Ji, Bo and Ye, Yangdong},
     journal = {IEEE Transactions on Systems, Man, and Cybernetics: Systems},
     doi = {10.1109/TSMC.2026.3663166},
     correspondingauthor = {Yangdong Ye},
     year = {2026}
   }
   

   This work studies full extrapolation over evolving temporal knowledge graphs, where entities, relations, and timestamps may all be unseen. The proposed contrastive variational graph symmetric autoencoder initializes unseen components from spatiotemporal relation connections and functional time encodings, then jointly encodes and reconstructs entity and relation representations. Dual contrastive objectives improve representation discrimination and task relevance, producing strong results on two benchmark datasets.

3. Jinlan Kong , Xiaohui He ^* , Haichuan Fang , Panle Li , Mengjia Qiao , Xijie Cheng , Haofei Li , Huitong Feng , Haonan Sun , Jiandong Shang

   MOUNT: Modality bottlenecked knowledge graph completion with multiple guidance exploitation

   Expert Systems with Applications , pp. 131174 (2026)

   DOI BIB Abstract

   @article{kong2026mount,
     title = {MOUNT: Modality bottlenecked knowledge graph completion with multiple guidance exploitation},
     author = {Kong, Jinlan and He, Xiaohui and Fang, Haichuan and Li, Panle and Qiao, Mengjia and Cheng, Xijie and Li, Haofei and Feng, Huitong and Sun, Haonan and Shang, Jiandong},
     journal = {Expert Systems with Applications},
     pages = {131174},
     doi = {10.1016/j.eswa.2026.131174},
     correspondingauthor = {Xiaohui He},
     year = {2026}
   }
   

   This work addresses redundant information within and across modalities in multimodal knowledge graph completion. MOUNT applies an information-bottleneck perspective with compression guidance to suppress intra-modal redundancy, alignment guidance to improve cross-modal consistency, and preservation guidance to retain task-relevant information. Joint optimization of these mechanisms yields robust entity representations and improves completion performance under both standard and noisy settings.

4. Qiang Guo , Bin Wu , Zhongchuan Sun , Haichuan Fang , Yangdong Ye ^*

   Knowledge-refined information bottleneck for contrastive recommendation

   Expert Systems with Applications , Vol. 294 , pp. 128673 (2025)

   DOI BIB Abstract

   @article{guo2025knowledge,
     title = {Knowledge-refined information bottleneck for contrastive recommendation},
     author = {Guo, Qiang and Wu, Bin and Sun, Zhongchuan and Fang, Haichuan and Ye, Yangdong},
     journal = {Expert Systems with Applications},
     volume = {294},
     pages = {128673},
     doi = {10.1016/j.eswa.2025.128673},
     correspondingauthor = {Yangdong Ye},
     year = {2025}
   }
   

   This work proposes a knowledge-refined information bottleneck framework for contrastive recommendation. An information-bottleneck regularized refinement component removes task-irrelevant knowledge graph information using dependence-based optimization, while a knowledge-integrated contrastive learning strategy preserves useful semantics across the user-item graph and knowledge graph. Experiments on four datasets show improved recommendation accuracy and resistance to noisy knowledge.

5. Yulin Du , Haichuan Fang , Hongfu Zuo ^* , Xincan Zhao ^*

   STF2: A novel spatial-temporal feature fusion method for aero-engine remaining useful life prediction

   Expert Systems with Applications , Vol. 290 , pp. 128370 (2025)

   DOI BIB Abstract

   @article{du2025stf2,
     title = {STF2: A novel spatial-temporal feature fusion method for aero-engine remaining useful life prediction},
     author = {Du, Yulin and Fang, Haichuan and Zuo, Hongfu and Zhao, Xincan},
     journal = {Expert Systems with Applications},
     volume = {290},
     pages = {128370},
     doi = {10.1016/j.eswa.2025.128370},
     correspondingauthor = {Hongfu Zuo|Xincan Zhao},
     year = {2025}
   }
   

   This work proposes a spatial-temporal feature fusion method for aircraft-engine remaining useful life prediction. Dynamic data augmentation combines Fourier and wavelet transformations to reduce noise, while parallel spatial-interaction and temporal-dependency modules model sensor correlations and degradation trends without sequential interference. A cross-attention fusion module integrates both feature types, and experiments on the C-MAPSS dataset demonstrate improved predictive performance.

6. Youwei Wang , Peisong Cao , Haichuan Fang , Yangdong Ye ^*

   Span-aware pre-trained network with deep information bottleneck for scientific entity relation extraction

   Neural Networks , Vol. 186 , pp. 107250 (2025)

   DOI BIB Abstract

   @article{wang2025span,
     title = {Span-aware pre-trained network with deep information bottleneck for scientific entity relation extraction},
     author = {Wang, Youwei and Cao, Peisong and Fang, Haichuan and Ye, Yangdong},
     journal = {Neural Networks},
     volume = {186},
     pages = {107250},
     doi = {10.1016/j.neunet.2025.107250},
     correspondingauthor = {Yangdong Ye},
     year = {2025}
   }
   

   This work presents a span-aware pretrained network with a deep information bottleneck for joint scientific entity and relation extraction. A span-based representation module separates task-relevant semantics from distracting context, while a task-relevant representation module captures dependencies across subtasks. An information minimum-maximum objective jointly compresses irrelevant information and strengthens shared predictive semantics, leading to improved results on scientific and biomedical datasets.

7. Haichuan Fang , Kexin Cheng , Ruixin Zhang , Youwei Wang , Yangdong Ye ^*

   Meta-collaboration-based semantic contrast for inductive knowledge representation learning

   Expert Systems with Applications , Vol. 261 , pp. 125421 (2025)

   DOI BIB Abstract

   @article{fang2025meta,
     title = {Meta-collaboration-based semantic contrast for inductive knowledge representation learning},
     author = {Fang, Haichuan and Cheng, Kexin and Zhang, Ruixin and Wang, Youwei and Ye, Yangdong},
     journal = {Expert Systems with Applications},
     volume = {261},
     pages = {125421},
     doi = {10.1016/j.eswa.2024.125421},
     correspondingauthor = {Yangdong Ye},
     year = {2025}
   }
   

   This work addresses semantic ambiguity caused by structural bias and the sparsity of newly observed entities in inductive knowledge representation learning. The proposed framework obtains relation-specific knowledge, aggregates neighborhood information with a multilayer graph neural network, and performs collaborative semantic contrast over support and query sets. Experiments on twelve inductive link-prediction benchmarks show that the learned representations are more discriminative and effective.

8. Zhen Tian , Yue Yu , Haichuan Fang , Weixin Xie , Maozu Guo ^*

   Predicting microbe–drug associations with structure-enhanced contrastive learning and self-paced negative sampling strategy

   Briefings in bioinformatics , Vol. 24 , No. 2 , pp. bbac634 (2023)

   DOI BIB Abstract

   @article{tian2023predicting,
     title = {Predicting microbe--drug associations with structure-enhanced contrastive learning and self-paced negative sampling strategy},
     author = {Tian, Zhen and Yu, Yue and Fang, Haichuan and Xie, Weixin and Guo, Maozu},
     journal = {Briefings in bioinformatics},
     volume = {24},
     number = {2},
     pages = {bbac634},
     doi = {10.1093/bib/bbac634},
     correspondingauthor = {Maozu Guo},
     year = {2023}
   }
   

   This work proposes a structure-enhanced contrastive learning and self-paced negative sampling method for predicting microbe-drug associations. Representations learned from meta-path-induced networks enhance embeddings from microbe and drug similarity networks through contrastive learning, while self-paced sampling selects informative negative examples for classifier training. Results on three public datasets and drug-focused case studies demonstrate improved association prediction.

9. Haichuan Fang , Youwei Wang , Zhen Tian ^* , Yangdong Ye ^*

   Learning knowledge graph embedding with a dual-attention embedding network

   Expert Systems with Applications , Vol. 212 , pp. 118806 (2023)

   DOI BIB Abstract

   @article{fang2023learning,
     title = {Learning knowledge graph embedding with a dual-attention embedding network},
     author = {Fang, Haichuan and Wang, Youwei and Tian, Zhen and Ye, Yangdong},
     journal = {Expert Systems with Applications},
     volume = {212},
     pages = {118806},
     doi = {10.1016/j.eswa.2022.118806},
     correspondingauthor = {Zhen Tian|Yangdong Ye},
     year = {2023}
   }
   

   This work proposes a dual-attention embedding network for knowledge graph representation learning. Bidirectional attention incorporates relation direction when aggregating neighboring entities, while relation-specific attention uses neighborhood evidence to update relation representations. Joint propagation enables entities and relations to interact semantically, and experiments on three standard link-prediction datasets show improvements over representative knowledge graph embedding baselines.

10. Zhen Tian , Xiangyu Peng , Haichuan Fang , Wenjie Zhang , Qiguo Dai , Yangdong Ye ^*

    MHADTI: predicting drug–target interactions via multiview heterogeneous information network embedding with hierarchical attention mechanisms

    Briefings in bioinformatics , Vol. 23 , No. 6 , pp. bbac434 (2022)

    DOI BIB Abstract

    @article{tian2022mhadti,
      title = {MHADTI: predicting drug--target interactions via multiview heterogeneous information network embedding with hierarchical attention mechanisms},
      author = {Tian, Zhen and Peng, Xiangyu and Fang, Haichuan and Zhang, Wenjie and Dai, Qiguo and Ye, Yangdong},
      journal = {Briefings in bioinformatics},
      volume = {23},
      number = {6},
      pages = {bbac434},
      doi = {10.1093/bib/bbac434},
      correspondingauthor = {Yangdong Ye},
      year = {2022}
    }
    

    This work introduces a multiview heterogeneous information network model with hierarchical attention for drug-target interaction prediction. The model constructs complementary network views, applies node-level and semantic-level attention to learn informative drug and target representations, and uses a neural decoder to estimate interaction probabilities. Evaluations across benchmark datasets show that integrating structural and semantic information from multiple views improves predictive performance.

11. Zhen Tian ^† , Haichuan Fang ^† , Zhixia Teng , Yangdong Ye ^*

    GOGCN: graph convolutional network on gene ontology for functional similarity analysis of genes

    IEEE/ACM Transactions on Computational Biology and Bioinformatics , Vol. 20 , No. 2 , pp. 1053–1064 (2022)

    DOI BIB Abstract

    @article{tian2022gogcn,
      title = {GOGCN: graph convolutional network on gene ontology for functional similarity analysis of genes},
      author = {Tian, Zhen and Fang, Haichuan and Teng, Zhixia and Ye, Yangdong},
      journal = {IEEE/ACM Transactions on Computational Biology and Bioinformatics},
      volume = {20},
      number = {2},
      pages = {1053--1064},
      year = {2022},
      doi = {10.1109/TCBB.2022.3181300},
      equalauthor = {Zhen Tian|Haichuan Fang},
      correspondingauthor = {Yangdong Ye}
    }
    

    This work presents GOGCN for measuring gene functional similarity by representing Gene Ontology as a graph. A graph-convolutional knowledge graph embedding model learns vector representations for ontology terms and relations, term similarity is computed from the learned embeddings, and a pair-wise strategy aggregates term similarities into gene-level functional similarity. Experiments on multiple datasets demonstrate improved reliability over established approaches.

Conference Proceedings

1. Haichuan Fang , Haoran Zhang , Yulin Du , Qiang Guo , Zhen Tian , Youwei Wang , Yangdong Ye ^*

   CDIB: consistency discovery-guided information bottleneck for multi-modal knowledge graph reasoning

   ACM International Conference on Multimedia , pp. 1062–1071 (2025)

   DOI BIB Abstract

   @inproceedings{fang2025cdib,
     title = {CDIB: consistency discovery-guided information bottleneck for multi-modal knowledge graph reasoning},
     author = {Fang, Haichuan and Zhang, Haoran and Du, Yulin and Guo, Qiang and Tian, Zhen and Wang, Youwei and Ye, Yangdong},
     booktitle = {ACM International Conference on Multimedia},
     pages = {1062--1071},
     doi = {10.1145/3746027.3754929},
     correspondingauthor = {Yangdong Ye},
     year = {2025}
   }
   

   This work develops a consistency discovery-guided information bottleneck framework for multimodal knowledge graph reasoning. A modality compression module reduces task-irrelevant information in individual modalities, a consistency discovery module identifies compatible cross-modal evidence during fusion, and an information preservation module retains predictive semantics. Evaluations on two benchmark datasets show competitive reasoning performance and robustness to noisy multimodal inputs.

2. Zhen Tian , Haichuan Fang , Yangdong Ye , Zhenfeng Zhu ^*

   SWE: a novel method with semantic-weighted edge for measuring gene functional similarity

   IEEE International Conference on Bioinformatics and Biomedicine , pp. 1672–1678 (2020)

   DOI BIB Abstract

   @inproceedings{tian2020swe,
     title = {SWE: a novel method with semantic-weighted edge for measuring gene functional similarity},
     author = {Tian, Zhen and Fang, Haichuan and Ye, Yangdong and Zhu, Zhenfeng},
     booktitle = {IEEE International Conference on Bioinformatics and Biomedicine},
     pages = {1672--1678},
     doi = {10.1109/BIBM49941.2020.9313495},
     correspondingauthor = {Zhenfeng Zhu},
     year = {2020}
   }
   

   This work proposes a semantic-weighted edge method for measuring gene functional similarity from the Gene Ontology graph. It estimates ontology-term semantics using structural information such as depth, ancestors, and descendants, assigns semantic contributions to graph relationships, and combines term-level evidence to compare genes. Experimental evaluations indicate that this structure-aware treatment improves functional similarity measurement over conventional information-content methods.