VGAT-VGAN Across Social Networks User Identity Linkage Algorithm Based on Fusion Features

doi:10.11871/jfdc.issn.2096-742X.2026.01.009

Abstract

Abstract:

[Purpose] The research on user identification across social networks is mainly to determine whether virtual users from different social networks belong to the same natural person. [Methods] To address the imbalance between positive and negative samples, firstly, the FD-Struc2vec and DW-Word2vec algorithms are proposed to extract the structural features of nodes and the text features of user names, respectively. Secondly, VGAT is used to optimize the structural feature representation, and the two types of features are fused to form a new user feature vector representation. Meanwhile, VGAN is used to increase the number of positive samples. Finally, Feature-MLP is proposed, which assigns different weights to structural features and text features in the neural network to realize user identification. [Results] Compared with the baseline algorithms such as WLAlign in real data sets, the results show that there is an improvement of more than 10% in the three indexes of P, R and F₁ value, which proves the effectiveness of the algorithm.[Limitations] Because social networks have a large number of users and complex friendships, coupled with the complexity of the algorithm structure, the overall computing demand is large, and the efficiency of the algorithm needs to be improved.

Key words: across social networks, user identity linkage, feature fusion, data enhancement, deep learning

PAN Yuquan,YUAN Deyu,JIA Yuan,WANG Anran. VGAT-VGAN Across Social Networks User Identity Linkage Algorithm Based on Fusion Features[J]. Frontiers of Data and Computing, 2026, 8(1): 103-118, https://cstr.cn/32002.14.jfdc.CN10-1649/TP.2026.01.009.

Figures/Tables 15

Fig.1

Fig.2

Algorithm 1

FD-Struc2vec algorithm"

算法1: FD-Struc2vec算法

Input: 社交网络关系图G.
Output: 各用户的节点特征向量.
Begin
将root添加到队列中，并标记为已访问
0 ← level
while (quene>0) and (level ≦ max_layers) do
count ← len(quene)
if reduce_len then
initialize deg_list 为字典
else
initialize deg_list 为列表
end if
whlie (count>0) do
for 每一个节点 do
从队列中取出节点，并获取度数
end for
if reduce_len then
update deg_list
else
将度数添加到deg_list
end if
for 邻居节点 do
if 邻居未访问 then
标记为已访问并加入队列
end if
end for
count ← count - 1
if reduce_len then
order_deg_list ← 将deg_list转换为有序列表
else
order_deg_list ← 将deg_list排序
end if
degree_seq ← order_deg_list
level ← level + 1
end while
end while
return degree_seq
for 每对结点(

v 1

, neighbors) do
获取

v 1

的度数列表list_1
for neighbors中的每个节点

v 2

do
获取

v 2

的度数列表list_2
计算有效层数max_layers
for layer from 0 to max_layers do
dtw_dist ← list_1[layer]和list_2[layer]的DTW距离
end for
end for
end for
return dtw_dist
//生成特征向量
execute biased_walk、random_work
return embedding
End

Algorithm 1

Algorithm 2

Fig.3

Fig.4

Table 1

Table 2

Table 3

Results of user identification model comparison"

身份关联算法	P	R	$F 1$
本文算法	0.8514	0.8326	0.8419
WLAlign	0.7952	0.7471	0.7381
PALE	0.6897	0.6382	0.6630
DeepLink	0.7736	0.7324	0.7547
CrossMNA	0.7598	0.6783	0.7167
B-GCN	0.8037	0.7859	0.7947
MNF-UI	0.7712	0.7852	0.7781

Table 3

Fig.5

Table 4

Results of node embedding algorithm comparison"

节点嵌入算法	P	R	$F 1$
FD-Struc2vec	0.7223	0.6854	0.7034
Struc2vec	0.6982	0.6521	0.6744
Node2vec	0.6728	0.6301	0.6508
LINE	0.6402	0.6007	0.6198
Deepwalk	0.6551	0.6298	0.6422

Table 4

Table5

Table 6

Experimental results of weight factor"

序号	a	b	P	R	$F 1$
1	0.55	0.45	0.7754	0.7458	0.7603
2	0.60	0.40	0.8121	0.7689	0.7899
3	0.65	0.35	0.8323	0.7854	0.8082
4	0.70	0.30	0.7556	0.7110	0.7326

Table 6

Table 7

Results of ablation experiment"

实验序号	VGAN	VGAT	Feature-MLP	P	R	$F 1$
1	-	-	-	0.6981	0.6521	0.6744
2	+	-	-	0.7224	0.6854	0.7034
3	-	+	-	0.7153	0.6758	0.6950
4	-	-	+	0.7301	0.7249	0.7275
5	+	-	+	0.8235	0.7762	0.7992
6	-	+	+	0.8027	0.7589	0.7802
7	+	+	-	0.7845	0.7271	0.7547
8	+	+	+	0.8514	0.8326	0.8419

Table 7

Fig.6

References 31

[1]	张津, 郭艳光. 基于偏好逻辑的社交网络用户身份识别方法[J]. 计算机仿真, 2022, 39(4): 450-453.
[2]	KEIKHA M M, MASEUD R, MASOUD A. DeepLink: A novel link prediction framework based on deep learning[J]. Journal Of Information Science,2021: 642-657.
[3]	王庚润. 网络空间用户身份对齐技术研究及应用综述[J]. 计算机科学, 2024, 51(5): 12-20. doi: 10.11896/jsjkx.230300172
[4]	戴军, 马强. 基于用户签到的跨社交网络用户匹配[J]. 计算机工程与应用, 2023, 59(2): 76-84. doi: 10.3778/j.issn.1002-8331.2203-0581
[5]	LI Y, PENG W, JI W, et al. User identification based on display names across online social networks[J]. IEEE Access, 2017, 5: 17342-17353. doi: 10.1109/ACCESS.2017.2744646
[6]	QU Y, XING L, MA H, et al. Exploiting User Friendship Networks for User Identification Across Social Networks[J]. Symmetry, 2022, 14(1): 110. doi: 10.3390/sym14010110
[7]	陈鸿昶, 徐乾, 黄瑞阳, 等. 一种基于用户轨迹的跨社交网络用户身份识别算法[J]. 电子与信息学报, 2018, 40(11): 2758-2764.
[8]	GAO X, JI W, LI Y, et al. User Identification with Spatiotemporal Awareness Across Social Networks[C]// Proceedings of the 27th ACM International Conference on Information and Knowledge Management, 2018.
[9]	CHEN W, WANG W, YIN H, et al. HFUL: A Hybrid Framework for User Account Linkage Across Location Aware Social Networks[J]. The VLDB Journal, 2023, 32(1): 1-22. doi: 10.1007/s00778-022-00730-8
[10]	RIEDERER C, KIM Y, CHAINTREAU A, et al. Linking Users Across Domains With Location Data: Theory And Validation[C]// Proceedings of the 25th international conference on world wide web, 2016.
[11]	HALIMI A, AYDAY E. Profile Matching Across Online Social Networks[C]// Information and Communications Security:22nd International Conference, ICICS 2020, Copenhagen, Denmark,Springer International Publishing, 2020.
[12]	ZHOU F, WEN Z, ZHONG T, et al. Unsupervised User Identity Linkage via Graph Neural Networks[C]// 2020 IEEE Global Communications Conference: GLOBECOM2020, Taipei, Taiwan, China, 2020:2975-2980.
[13]	ABBAS, ASH MOHAMMAD. Social Network Analysis Using Deep Learning: ApplicationsAnd Schemes[J]. Social Network Analysis and Mining, 2021, 11(1): 106. doi: 10.1007/s13278-021-00799-z
[14]	LIU L, CHEN P, LI X, et al. Wlalign: Weisfeiler-lehman relabeling for aligning users across networks via regularized representation learning[J]. IEEE Transactions on Knowledge andData Engineering, 2023, 36(1): 445-458.
[15]	CHEN H, YIN H, SUN X, et al. Multi-levelgraph convolutional networks for cross-platform anchor link prediction[C]// Proceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining. 2020.
[16]	SHEN Y, JIANG X, LI Z, et al. UniSKGRep: A unified representation learning framework of social network and knowledge graph[J]. Neural Networks, 2023, 158: 142-153. doi: 10.1016/j.neunet.2022.11.010
[17]	ZHOU F, LIU L, ZHANG K, et al. Deeplink: A deep learning approach for user identity linkage[C]// IEEE INFOCOM 2018-IEEE Conference on computer communications. IEEE, 2018.
[18]	TRUNG H T, VAN VINH T, TAM N T, et al. Adaptive network alignment with unsupervised and multiorder convolutional networks[C]// 2020 IEEE 36th International Conference onData Engineering (ICDE). IEEE, 2020.
[19]	LI X, CAO Y, LI Q, et al. RLINK: Deep Reinforcement Learning For User Identity Linkage[J]. World Wide Web, 2021, 24: 85-103. doi: 10.1007/s11280-020-00833-8
[20]	MAN T, SHEN H, LIU S, et al. Predict anchor links across social networks via an embedding approach[C]// Ijcai. 2016, 16: 1823-1829.
[21]	CHU X, FAN X, YAO D, et al. Cross-network embedding for multi-network alignment[C]// In The worldwide web conference, 2019: 273-284.
[22]	ZHOU X, LIANG X, DU X, et al. Structure based user identification across social networks[J]. IEEE Transactions on Knowledge and Data Engineering, 2017, 30(6): 1178-1191. doi: 10.1109/TKDE.69
[23]	HUANG S, XIANG H, LENG C, et al. Cross-Social-Network User Identification Based on Bidirectional GCN and MNF-UI Models[J]. Electronics 2024, 13, 2351.
[24]	CHEN W, WANG W, YIN H, et al. User Account Linkage Across Multiple Platforms With Location Data[J]. Journal of Computer Science and Technology, 2020, 35: 751-768. doi: 10.1007/s11390-020-0250-7
[25]	袁烽. 跨社交网络用户身份识别关键技术研究[D]. 南京: 东南大学, 2022.
[26]	QI M, WANG Z, HE Z, et al. User Identification Across Asynchronous Mobility Trajectories[J]. Sensors, 2019, 19(9): 2102. doi: 10.3390/s19092102
[27]	LI X, SHANG Y, CAO Y, et al. Type-Aware Anchor Link Prediction Across Heterogeneous Networks Based On Graph Attention Network[C]// Proceedings of the AAAI Conference on Artificial Intelligence, 2020.
[28]	DING X, ZHANG H, MA C, et al. User Identification Across Multiple Social Networks Based On Naive Bayes Model[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022, 35(3): 4274-4285. doi: 10.1109/TNNLS.2022.3202709
[29]	吴劲, 陈树沛, 杨庆, 等. 基于图神经网络的用户轨迹分类[J]. 电子科技大学学报, 2021, 50(5): 734-740.
[30]	YANG J, ZHOU W, QIAN H, et al. Topic Sequence Embedding for User Identity Linkage from Heterogeneous Behavior Data[C].// ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing. IEEE, 2021: 2590-2594.
[31]	ZHANG Y, TANG J, YANG Z, et al. COSNET: Connecting Heterogeneous Social Networks with Local and Global Consistency[C]// Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 2015: 1485-1494.

状态	社交网络类型	节点数	边数	关联节点数	无关节点数	比值
处理前	Flickr	215,495	9,114,557	641	214,854	335.1856
处理前	Lastfm	136,420	1,685,524	641	135,957	212.1014
处理后	Flickr	171,028	1,497,165	452	170,576	377.3805
处理后	Lastfm	136,409	1,685,469	452	135,779	300.3960

序号	Flickr用户名	Lastfm用户名	相似类别
1	karles	smkn	完全不相同
2	type_girl	type_girl	完全相同
3	lcfik	LCFik	改变大小写
4	griffey/jhench1638	jgriffey/jhench	添加字符串前缀/后缀
5	apple	^apple^	加入特殊符号
6	clintonjeff	cj	缩写
7	-_-_-_-	=_=_=_=	无法识别，需要替换

用户名来源（同一自然人）		嵌入算法结果的相似度
Flickr	Lastfm	DW-Word2vec	Word2vec	GloVe	BERT
ine_ber	ineber	0.8798	0.7812	0.6945	0.7461
marcogomes	MARcoGomes	0.8481	0.7680	0.7235	0.7656
IPDpl	IpD*pk	0.7471	0.6308	0.6412	0.6190
Pvw2180	pvw	0.6249	0.5756	0.6188	0.5873
editor	editor	0.9306	0.8745	0.8529	0.8223