融合知识增强和多层抑噪的个性化推荐模型

谢珺; 王丹彤; 张博; 陈桂军; 吕佳琪; 雒雄艳

doi:10.11999/JEIT260051

融合知识增强和多层抑噪的个性化推荐模型

doi: 10.11999/JEIT260051 cstr: 32379.14.JEIT260051

太原理工大学电子信息工程学院山西 030600

基金项目: 国家自然科学基金(62201377)，虚拟现实技术与系统国家重点实验室(北京航空航天大学)开放课题基金(VRLAB2022C11)，山西省回国留学项目(2024-61)

详细信息

作者简介:
谢珺：女，副教授，研究方向为智能信息处理、知识图谱理论和推荐系统

王丹彤：男，硕士生，研究方向为知识图谱、推荐系统

张博：男，副教授，研究方向为具身智能和智能信息处理

陈桂军：男，副教授，研究方向为智能信息处理和脑机接口

吕佳琪：女，硕士生，研究方向为知识图谱和人工智能

雒雄艳：女，硕士生，研究方向为大语言模型和推荐系统

通讯作者:
张博　zhangbo@tyut.edu.cn

中图分类号: TP391.3
计量
- 文章访问数: 76
- HTML全文浏览量: 24
- PDF下载量: 13
- 被引次数: 0
出版历程
- 收稿日期: 2026-01-14
- 修回日期: 2026-04-28
- 录用日期: 2026-05-12
- 网络出版日期: 2026-06-02

KE-HNS: Knowledge-Enhanced Personalized Recommendation Model with Hierarchical Noise Suppression

School of Electronic and Information Engineering, Taiyuan University of Technology, Jinzhong 030600, China

Funds: The National Natural Science Foundation of China(62201377), The Open Project Program of State Key Laboratory of Virtual Reality Technology and Systems (VRLAB2022C11), Shanxi Scholarship Council of China (2024-61)

摘要

摘要: 现有的基于知识图谱的推荐模型面临着噪声抑制不足、细粒度用户兴趣捕捉不充分以及信息利用不平衡等挑战，影响推荐的准确性。该文提出一种融合知识增强和多层抑噪的个性化推荐模型(KE-HNS)。首先，提出一种多层抑噪策略：在输入端构建降噪层，分别以2元掩码矩阵与掩盖低权重3元组的方式，剔除用户历史交互行为及知识图谱中的无关或误导性信息；在主干网络上设置抑噪层，利用实体关系空间隔离消除特征融合过程中的语义污染；同时构造对比学习压噪层，通过正负样本的比较压制无关实体噪声。其次，为更精准建模用户偏好与项目特性，利用图卷积网络、自适应权重层，借助降噪后的外部知识增强用户表征和项目表征。再次，通过构建用户-项目与项目-实体双视图对比学习机制，KE-HNS借助正负样本对自适应地平衡不同信息源的权重，从而缓解信息利用不平衡的问题。在3个公开基准数据集上的实验表明，与(CG-KGR)、(KGIN)和(KG-DCRec)等多种先进方法对比，所提模型KE-HNS在AUC和F1指标上分别提升了0.94%～1.01%和0.43%～0.90%，在降噪性能对比和Recall@K等指标上也优于基准模型。
- 推荐系统 /
- 图神经网络 /
- 多层抑噪 /
- 知识增强 /
- 对比学习
Abstract: Objective In the era of Big Data and Artificial Intelligence (AI), rapid information growth has increased the difficulty of filtering valuable content from redundant data. Personalized recommender systems are key tools for accurate information matching and resource allocation. Knowledge Graphs (KGs) can enrich user-item representations. However, current KG-based recommendation models still face weak noise suppression, coarse-grained user-interest modeling, and imbalanced use of heterogeneous information, which reduce recommendation accuracy. This paper proposes Knowledge-Enhancedpersonalized recommender model with HierarchicalNoise Suppression (KE-HNS), which integrates knowledge enhancement with hierarchical noise suppression. By combining graph representation learning and contrastive learning, KE-HNS addresses noise interference, fine-grained preference modeling, and multi-source information balance, thereby improving recommendation performance. Methods KE-HNS adopts a hierarchical noise-suppression paradigm. At the input stage, Input Noise Reduction (INR) is used to reduce noise from two sources. For user-item interactions, a learnable binary mask matrix is used to remove noisy edges. For KG denoising enhancement, triples are scored by importance, low-score triples are identified with a Bottom-K strategy, and noisy triples are masked. At the feature-fusion stage, Isolated Noise Suppression (INS) is used to preserve spatial independence by partitioning entity-attribute spaces according to relation type. This design limits high-order noise propagation and semantic contamination. At the representation-optimization stage, Comparative Noise Suppression (CNS) is implemented through contrastive learning to suppress irrelevant entity noise and strengthen robust semantic signals. To capture fine-grained user interests, Graph Convolutional Networks (GCNs) are used to enhance user representations from historical interactions and related entities. Adaptive weight layers further refine item representations by using entity attributes and relations. To balance heterogeneous information, a dual-view contrastive learning mechanism is constructed between the user-item view and the item-entity view. Positive and negative sample pairs are used to adaptively adjust the weights of different information sources. Finally, user and item representations are matched by inner product to generate the Top-K recommendation list. Results and Discussions KE-HNS is evaluated on three public datasets, Book-Crossing, MovieLens-1M, and Last.FM, through performance comparison, ablation experiments, denoising evaluation, case analysis, and complexity assessment. For Click-Through Rate (CTR) prediction, KE-HNS outperforms the best baseline models by 0.94%–1.01% in Area Under the Curve (AUC) and 0.43%～0.90% in F1-score (Table 3). For Top-K recommendation, its Recall@K is higher than those of most advanced methods across nearly all K values, with only a slight gap behind CG-KGR on Last.FM (Fig. 7). The ablation results show that all three denoising components contribute to the performance gains (Table 4). The denoising evaluation shows that KE-HNS effectively suppresses noise and maintains high prediction accuracy under noisy conditions (Fig. 8). The complexity analysis further indicates that the model remains feasible for practical deployment (Table 5). Conclusions This paper presents KE-HNS, a personalized recommendation model that combines knowledge enhancement with hierarchical noise suppression. By reducing noise interference and balancing collaborative filtering signals with knowledge-aware semantics, KE-HNS improves recommendation accuracy across multiple benchmark datasets. The model still has limitations in computational efficiency and depends on the coverage and completeness of the KG. Future work may focus on computational optimization and dynamic knowledge integration.
- Recommender system /
- Graph neural network /
- Hierarchical noise suppression /
- Knowledge enhancement /
- Contrastive learning

HTML全文

图 1 大数据时代下知识图谱赋能推荐系统

下载: 全尺寸图片幻灯片

图 2 KE-HNS模型框架示意图

下载: 全尺寸图片幻灯片

图 3 模块作用及多层降噪体现

下载: 全尺寸图片幻灯片

图 4 用户-项目交互降噪机制

下载: 全尺寸图片幻灯片

图 5 知识图降噪增强机制

下载: 全尺寸图片幻灯片

图 6 属性空间隔离建模过程

下载: 全尺寸图片幻灯片

图 7 Top-K推荐任务结果

下载: 全尺寸图片幻灯片

图 8 降噪性能对比

下载: 全尺寸图片幻灯片

图 9 KE-HNS模型在推荐系统领域有效性的例证

下载: 全尺寸图片幻灯片

表 1 数据集统计

数据集	知识图谱			用户项目交互
数据集	实体	关系	3元组	用户	项目	交互
MovieLens-1M	182011	12	1241996	6036	2445	753772
Last.FM	9366	60	15518	1872	3846	42346
Book-Crossing	77903	25	151500	17860	14976	139746

下载: 导出CSV

表 2 基线模型简介

基线模型	时间	简介
CKE	2016	一种融合知识图谱与协同过滤的推荐系统模型，旨在整合多源异构信息。
RippleNet	2018	一种通过知识图谱语义关联将用户历史兴趣传播至候选项目的推荐系统模型。
KGAT	2019	一种采用协同知识图谱显式捕获交互与图谱间的高阶语义关系的推荐模型。
KGIN	2021	一种融合细粒度用户意图建模与关系路径感知聚合的推荐系统模型。
CG-KGR	2022	一种通过协同引导机制融合用户-项目交互数据与外部知识图谱的推荐模型。
MDCLBR	2024	一种基于用户-项目簇、用户-项目及项目簇-项目三视角多视图对比学习模型。
KG-DCRec	2025	一种融合差异化去噪与双对比学习的图神经网络推荐模型。

下载: 导出CSV

表 3 AUC和F1结果对比

模型	MovieLens-1M		Last.FM		Book-Crossing
模型	AUC	F1	AUC	F1	AUC	F1
CKE	0.9066	0.8032	0.7471	0.6743	0.6762	0.6235
RippleNet	0.9188	0.8420	0.7762	0.7025	0.7217	0.6465
KGAT	0.9110	0.8426	0.8292	0.7424	0.7314	0.6547
CG-KGR	0.9114	0.8422	0.8493	0.7527	0.7419	0.6540
KGIN	0.9190	0.8440	0.8486	0.7602	0.7265	0.6619
MDCLBR	0.9213	0.8438	0.8502	0.7593	0.7506	0.6642
KG-DCRec	0.9261	0.8547	0.8477	0.7706	0.7541	0.6674
KE-HNS	0.9349	0.8624	0.8588	0.7739	0.7612	0.6731
IMPROVE(%)	0.95	0.90	1.01	0.43	0.94	0.85

下载: 导出CSV

表 4 消融实验结果

变体模型	AUC	F1
w/o INR&INS&CNS	0.9005	0.8268
w/o INR&INS	0.9058	0.8302
w/o INR&CNS	0.9040	0.8317
w/o INS&CNS	0.9192	0.8465
w/o INR	0.9208	0.8483
w/o CNS	0.9293	0.8598
w/o INS	0.9301	0.8613
KE-HNS	0.9349	0.8624

下载: 导出CSV

表 5 算法复杂度和训练时间对比

模型	计算复杂度	训练时间(s)
MDCLBR	O(L(2\|V\|Ed+E\|d)+\|V\|²log\|V\| +2\|V\|²d+k\|V\|d+\|N\|d)	5.404
CG-KGR	O(L(\|V\|d²+2\|N\|d)+\|V\|d²+\|N\|d)	2.2147
KGIN	O(d^L+\|V\|d²+2\|E\|d +\|V\|d+\|N\|d)	506.8
KE-HNS	O(L(kd²+\|U\|\|\|I\|)²d+2\|E\|kd) +\|E\|Md+2\|V\|²d)	7.548

下载: 导出CSV

参考文献(33)

[1]	韩一鸣, 魏国辉. 知识图谱与推荐算法任务优化及领域应用发展综述[J]. 计算机工程与应用, 2026, 62(10): 26–53. doi: 10.3778/j.issn.1002-8331.2505-0279. HAN Yiming and WEI Guohui. Overview of task optimization and domain application development of knowledge graph and recommendation algorithm[J]. Computer Engineering and Applications, 2026, 62(10): 26–53. doi: 10.3778/j.issn.1002-8331.2505-0279.
[2]	IBRAHIM O A S, YOUNIS E M G, MOHAMED E A, et al. Revisiting recommender systems: An investigative survey[J]. Neural Computing and Applications, 2025, 37(4): 2145–2173. doi: 10.1007/s00521-024-10828-5.
[3]	RAJAL R, KUMAR N, KUMAR S, et al. A comprehensive review of product recommendation systems using deep learning techniques[J]. Knowledge and Information Systems, 2026, 68(1): 8. doi: 10.1007/s10115-025-02667-0.
[4]	GARAPATI R and CHAKRABORTY M. Recommender systems in the digital age: A comprehensive review of methods, challenges, and applications[J]. Knowledge and Information Systems, 2025, 67(8): 6367–6411. doi: 10.1007/s10115-025-02453-y.
[5]	高云君, 葛丛丛, 郭宇翔, 等. 面向关系型数据与知识图谱的数据集成技术综述[J]. 软件学报, 2023, 34(5): 2365–2391. doi: 10.13328/j.cnki.jos.006808. GAO Yunjun, GE Congcong, GUO Yuxiang, et al. Survey on data integration technologies for relational data and knowledge graph[J]. Journal of Software, 2023, 34(5): 2365–2391. doi: 10.13328/j.cnki.jos.006808.
[6]	廖涛, 张国畅, 张顺香. 基于双粒度图的文档级关系抽取[J]. 计算机工程, 2024, 50(10): 164–173. doi: 10.19678/j.issn.1000-3428.0068728. LIAO Tao, ZHANG Guochang, and ZHANG Shunxiang. Document-level relation extraction based on dual-granularity graphs[J]. Computer Engineering, 2024, 50(10): 164–173. doi: 10.19678/j.issn.1000-3428.0068728.
[7]	WANG Binhao, XIAO Yutian, WANG Maolin, et al. SPARK: Adaptive low-rank knowledge graph modeling in hybrid geometric spaces for recommendation[C]. The 34th ACM International Conference on Information and Knowledge Management, Seoul, Republic of Korea, 2025: 2956–2966. doi: 10.1145/3746252.3760999.
[8]	LI Haobin, LIN Yijie, HU Peng, et al. Learning with dual-level noisy correspondence for multi-modal entity alignment[C]. The 14th International Conference on Learning Representations, Rio de Janeiro, Brazil, 2026.
[9]	刘超, 余岩化. 融合降噪策略与多视图对比学习的知识感知推荐模型[J]. 计算机应用, 2025, 45(9): 2827–2837. doi: 10.11772/j.issn.1001-9081.2024081225. LIU Chao and YU Yanhua. Knowledge-aware recommendation model combining denoising strategy and multi-view contrastive learning[J]. Journal of Computer Applications, 2025, 45(9): 2827–2837. doi: 10.11772/j.issn.1001-9081.2024081225.
[10]	CHEN Yankai, YANG Yaming, WANG Yujing, et al. Attentive knowledge-aware graph convolutional networks with collaborative guidance for personalized recommendation[C]. IEEE 38th International Conference on Data Engineering, Kuala Lumpur, Malaysia, 2022: 299–311. doi: 10.1109/ICDE53745.2022.00027.
[11]	SANG Lei, HU Yang, ZHANG Yi, et al. Multi-view denoising contrastive learning for bundle recommendation[J]. Applied Intelligence, 2024, 54(23): 12332–12346. doi: 10.1007/s10489-024-05825-z.
[12]	WANG Hui, LI Qin, LUO Huilan, et al. The graph attention recommendation method for enhancing user features based on knowledge graphs[J]. Mathematics, 2025, 13(3): 390. doi: 10.3390/math13030390.
[13]	吴浩萌. 推荐系统中的深度矩阵分解方法研究及应用[D]. [硕士论文], 吉林大学, 2022. doi: 10.27162/d.cnki.gjlin.2022.006200. WU Haomeng. Research and application of deep matrix factorization method in recommender system[D]. [Master dissertation], Jilin University, 2022. doi: 10.27162/d.cnki.gjlin.2022.006200.
[14]	HE Xiangnan, LIAO Lizi, ZHANG Hanwang, et al. Neural collaborative filtering[C]. The 26th International Conference on World Wide Web, Perth, Australia, 2017: 173–182. doi: 10.1145/3038912.3052569.
[15]	LIU Yiqun and ZHANG Min. TikTok recommendation algorithms[J]. Science Foundation in China, 35(3): 415–416. doi: 10.16262/j.cnki.1000-8217.2021.03.021.
[16]	YAO Yuan, ZHAN Hui, NOORIAN A, et al. Enhancing POI recommendations on social media: A sequential approach incorporating LSTM and user feedback[J]. Computing, 2025, 107(1): 21. doi: 10.1007/s00607-024-01385-9.
[17]	PENG Jinglin and ZHOU Jianfeng. Framework for building knowledge map of ethnic music based on big data[J]. Journal of Electronics and Information Science, 2024, 9(1): 69–79. doi: 10.23977/jeis.2024.090111.
[18]	陈玲洪, 潘晓华. 基于知识图谱和读者画像的图书推荐研究[J]. 数据分析与知识发现, 2023, 7(12): 164–171. doi: 10.11925/infotech.2096-3467.2022.1065. CHEN Linghong and PAN Xiaohua. Recommending books based on knowledge graph and reader profiling[J]. Data Analysis and Knowledge Discovery, 2023, 7(12): 164–171. doi: 10.11925/infotech.2096-3467.2022.1065.
[19]	ZHANG Yiwen, ZHANG Li, DONG Yunchun, et al. A movie recommendation method based on knowledge graph and time series[J]. Journal of Intelligent & Fuzzy Systems, 2023, 45(3): 4715–4724. doi: 10.3233/JIFS-230795.
[20]	程开原, 姚俊萍, 李晓军, 等. 时态网络中知识图谱推荐: 关键技术与研究进展[J]. 中国电子科学研究院学报, 2021, 16(2): 174–183,188. doi: 10.3969/j.issn.1673-5692.2021.02.011. CHENG Kaiyuan, YAO Junping, LI Xiaojun, et al. Recommendation based on knowledge graph in temporal networks: Key technologies and progress[J]. Journal of China Academy of Electronics and Information Technology, 2021, 16(2): 174–183,188. doi: 10.3969/j.issn.1673-5692.2021.02.011.
[21]	ZHANG Fuzheng, YUAN N J, LIAN Defu, et al. Collaborative knowledge base embedding for recommender systems[C]. The 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Francisco, USA, 2016: 353–362. doi: 10.1145/2939672.2939673.
[22]	WANG Hongwei, ZHANG Fuzheng, WANG Jialin, et al. RippleNet: Propagating user preferences on the knowledge graph for recommender systems[C]. The 27th ACM International Conference on Information and Knowledge Management, Torino, Italy, 2018: 417–426. doi: 10.1145/3269206.3271739.
[23]	WANG Hongwei, ZHANG Fuzheng, XIE Xing, et al. DKN: Deep knowledge-aware network for news recommendation[C]. The 2018 World Wide Web Conference, Lyon, France, 2018: 1835–1844. doi: 10.1145/3178876.3186175.
[24]	YU Xiao, REN Xiang, SUN Yizhou, et al. Personalized entity recommendation: A heterogeneous information network approach[C]. The 7th ACM International Conference on Web Search and Data Mining, New York, USA, 2014: 283–292. doi: 10.1145/2556195.2556259.
[25]	WANG Xiang, WANG Dingxian, XU Canran, et al. Explainable reasoning over knowledge graphs for recommendation[C]. The 33rd AAAI Conference on Artificial Intelligence, Honolulu, USA, 2019: 5329–5336. doi: 10.1609/aaai.v33i01.33015329.
[26]	ZHAO Huan, YAO Quanming, LI Jianda, et al. Meta-graph based recommendation fusion over heterogeneous information networks[C]. The 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Halifax, Canada, 2017: 635–644. doi: 10.1145/3097983.3098063.
[27]	YANG Yuhao, HUANG Chao, XIA Lianghao, et al. Knowledge graph contrastive learning for recommendation[C]. The 45th International ACM SIGIR Conference on Research and Development in Information Retrieval, Madrid, Spain, 2022: 1434–1443. doi: 10.1145/3477495.3532009.
[28]	ZHOU Jingyan, SHAN Zhilong, WU Zhengyang, et al. Knowledge graph denoising with dual contrast for recommendation[C]. 21st International Conference on Advanced Intelligent Computing Technology and Applications, Ningbo, China, 2025: 350–361. doi: 10.1007/978-981-95-0009-3_30.
[29]	SU Jiajie, ZHENG Xiaolin, LIN Zibin, et al. Contrastive hypergraph flows with multifaceted gates for cross-domain sequential recommendation[J]. IEEE Transactions on Services Computing, 2025, 18(6): 3595–3608. doi: 10.1109/TSC.2025.3620425.
[30]	SUN Shengyin and MA Chen. Hyperbolic contrastive learning with model-augmentation for knowledge-aware recommendation[C]. Joint European Conference on Machine Learning and Knowledge Discovery in Databases, Vilnius, Lithuania, 2024: 199–217. doi: 10.1007/978-3-031-70371-3_12.
[31]	WANG Xiang, HE Xiangnan, CAO Yixin, et al. KGAT: Knowledge graph attention network for recommendation[C]. The 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, Anchorage, USA, 2019: 950–958. doi: 10.1145/3292500.3330989.
[32]	WANG Xiang, HUANG Tinglin, WANG Dingxian, et al. Learning intents behind interactions with knowledge graph for recommendation[C]. The Web Conference 2021, Ljubljana, Slovenia, 2021: 878–887. doi: 10.1145/3442381.3450133.
[33]	ZHAO W X, MU Shanlei, HOU Yupeng, et al. RecBole: Towards a unified, comprehensive and efficient framework for recommendation algorithms[C]. The 30th ACM International Conference on Information & Knowledge Management, 2021: 4653–4664. doi: 10.1145/3459637.3482016.