Recommendation Algorithm Based on Social Diffusion and Adaptive Negative Sampling

doi:10.12141/j.issn.1000-565X.250179

Abstract

Abstract:

Social recommendation algorithms based on Graph Neural Network (GNN) leverage social networks to improve recommendation performance. However, most existing methods directly integrate the raw social graph into the recommendation system, which often introduces noise as they overlook the presence of non-homophilous social connections. Furthermore, prevailing negative sampling strategies typically select negative samples with a fixed level of hardness, which is prone to generating false negatives and consequently limits the model’s ability to effectively discriminate between user preferences. To address these issues, this paper proposed a novel recommendation algorithm based on social diffusion and adaptive negative sampling. First, forward diffusion and interest-guided denoising were performed on the social network to derive user representations that reflect homophilic social relations. Subsequently, a multi-view representation alignment approach was employed to maximize the mutual information among user representations from the denoised social graph, the original social graph, and the user-item interaction graph, thereby enhancing the quality of user embeddings. Finally, negative samples of adaptive hardness were selected based on the predicted scores of positive samples, enabling dynamic calibration of the similarity boundary between positive and negative pairs to improve overall model performance. Extensive experimental results demonstrate that the proposed algorithm significantly outperforms state-of-the-art recommendation baselines. On the Douban dataset, it improves recall and NDCG by 11.99% and 10.54%, respectively; on Epinions, by 15.62% and 11.14%; and on Yelp, by 13.80% and 14.90%. These results validate its effectiveness in alleviating social noise and enhancing the differentiation between positive and negative samples.

Key words: recommendation algorithm, social network, graph neural network, diffusion model, contrastive learning, negative sampling

CLC Number:

TP391

CAI Xiaodong, LI Ting, SU Yifeng. Recommendation Algorithm Based on Social Diffusion and Adaptive Negative Sampling[J]. Journal of South China University of Technology(Natural Science Edition), 2026, 54(2): 52-61.

Figures/Tables 13

Fig.1

Fig.2

Fig.3

Fig.4

Table 1

Table 2

Fig.5

Table 3

Fig.6

Fig.7

Table 4

Fig.8

Fig.9

References 20

[1]	蔡晓东，周青松，叶青．基于动态邻域采样的社交推荐模型［J］．华南理工大学学报（自然科学版），2024，52（2）：32-41.
	CAI Xiaodong， ZHOU Qingsong， YE Qing ．Social reco-mmendation model based on dynamic neighborhood sampling［J］．Journal of South China University of Techno-logy （Natural Science Edition），2024，52（2）：32-41.
[2]	MCPHERSON M， SMITH L， COOK J M ．Birds of a feather：homophily in social networks［J］．Annual Review of Sociology，2001，27（1）：415-444.
[3]	SHARMA K， LEE Y C， NAMBI S，et al ．A survey of graph neural networks for social recommender systems ［J］．ACM Computing Survey，2024，56（10）：1-34.
[4]	ANAND V， MAURYA A K ．A survey on recommender systems using graph neural network［J］．ACM Transactions on Information Systems，2025，43（1）：1-49.
[5]	LIAO J， ZHOU W， LUO F，et al ．SocialLGN：light graph convolution network for social recommendation［J］．Information Sciences，2022，589（15）：595-607.
[6]	XIA L， SHAO Y， HUANG C，et al ．Disentangled graph social recommendation［C］∥ Proceedings of 2023 IEEE the 39th International Conference on Data Engineering．Anaheim：IEEE，2023：2332-2344.
[7]	YU Z， WANG Y， HU M，et al ．Integrating social and knowledge graphs in GNN-based recommender systems ［C］∥ Proceedings of 2024 International Joint Conference on Neural Networks．Yokohama：IEEE，2024：1-8.
[8]	JIANG W， GAO X， XU G，et al ．Challenging low homophily in social recommendation［C］∥ Proceedings of the ACM Web Conference 2024．Singapore：ACM，2024：3476-3484.
[9]	HU Z， NAKAGAWA S， ZHUANG Y，et al ．Hierarchical denoising for robust social recommendation［J］．IEEE Transactions on Knowledge and Data Engineering，2025， 37（2）：739-753.
[10]	QUAN Y， DING J， GAO C ．Robust preference-guided denoising for graph based social recommendation ［C］∥ Proceedings of the ACM Web Conference 2023．Austin：ACM，2023：1097-1108.
[11]	MA W， WANG Y， ZHU Y，et al ．MADM：a modela-gnostic denoising module for graph-based social recommendation［C］∥ Proceedings of the 17th ACM Inter-national Conference on Web Search and Data Mining．Merida：ACM，2024：501-509.
[12]	YANG Y， WU L， WANG Z，et al ．Graph bottlenecked social recommendation［C］∥ Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining．Barcelona：ACM，2024：3853-3862.
[13]	LAI R， CHEN L， ZHAO Y，et al ．Disentangled negative sampling for collaborative filtering［C］∥ Proceedings of the 16th ACM International Conference on Web Search and Data Mining．Singapore：ACM，2023：96-104.
[14]	ZHAO Y， CHEN R， LAI R，et al ．Augmented negative sampling for collaborative filtering［C］∥ Procee-dings of the 17th ACM Conference on Recommender Systems．Singapore： ACM，2023：256-266.
[15]	马晓亮，高洁，刘英，等．基于意图理解驱动的客服知识推荐大模型构建［J］．华南理工大学学报（自然科学版），2025，53（3）：40-49.
	MA Xiaoliang， GAO Jie， LIU Ying，et al ．Customer service knowledge recommendation large model construction driven by intent understanding［J］．Journal of South China University of Technology （Natural Science Edition），2025，53（3）：40-49.
[16]	ZHAO J， WANG W， XU Y，et al ．Denoising diffusion recommender model［C］∥ Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval．Washington：ACM，2024：1370-1379.
[17]	LI A， BO Y ．Dual graph denoising model for social recommendation［C］∥ Proceedings of the ACM on Web Conference 2025．Sydney：ACM，2025：347-356.
[18]	SUN Y， SUN Z， DU Y，et al ．Model-agnostic social network refinement with diffusion models for robust social recommendation［C］∥ Proceedings of the ACM on Web Conference 2025．Sydney：ACM，2025：370-378.
[19]	YANG Z， DING M， HUNAG T，et al ．Does negative sample matter？A review with insights into its theory and applications［J］．IEEE Transactions on Pattern Analysis and Machine Intelligence，2024，46（8）：5692-5710.
[20]	WANG T， XIA L， HUANG C，et al ．Denoised self-augmented learning for social recommendation［C］∥ Proceedings of the 32th International Joint Conference on Artificial Intelligence．San Francisco：Morgan Kaufmann，2023：2324-2331.

数据集	数量				稠密度/%
数据集	用户	项目	交互	社交	交互	社交
Douban	12 910	22 291	591 296	169 150	0.21	0.10
Epinions	22 160	296 269	922 116	355 754	0.01	0.07
Yelp	30 934	22 123	441 304	1 001 010	0.06	0.10

模型	R@20			NDCG@20
模型	Douban	Epinions	Yelp	Douban	Epinions	Yelp
GDMSR	0.140 9	0.039 1	0.096 3	0.094 8	0.023 6	0.050 6
DSL	0.127 9	0.052 4	0.100 4	0.083 3	0.031 3	0.054 9
SHaRe	0.146 5	0.048 0	0.093 3	0.097 3	0.027 8	0.049 3
GBSR	0.165 5	0.054 4	0.115 7	0.115 1	0.032 0	0.063 4
HDSR	0.166 8	0.060 8	0.119 6	0.115 7	0.035 9	0.065 1
SARec	0.186 8	0.070 3	0.136 1	0.127 9	0.039 9	0.074 8

λ₂	R@20	NDCG@20
0.001	0.157 2	0.101 7
0.010	0.164 8	0.107 5
0.050	0.186 9	0.128 3
0.100	0.181 5	0.124 9
0.500	0.122 6	0.069 2

模型	训练时间/s			测试时间/s
模型	Douban	Epinions	Yelp	Douban	Epinions	Yelp
DSL	36.04	72.57	21.08	3.02	5.15	2.56
GBSR	24.92	44.75	21.92	2.50	4.65	1.85
HDSR	23.38	43.96	13.90	3.95	7.58	2.63
SARec	39.72	74.41	25.24	4.78	8.18	3.31