CAT: Can Trust be Predicted with Context-Awareness in Dynamic Heterogeneous Networks?

Jie Wang; Zheng Yan; Jiahe Lan; Xuyan Li; Elisa Bertino

cs.LGDec 12, 2025

CAT: Can Trust be Predicted with Context-Awareness in Dynamic Heterogeneous Networks?

Authors: Jie Wang, Zheng Yan, Jiahe Lan, Xuyan Li, Elisa Bertino

Organizations: State Key Laboratory of Integrated Services Networks, School of Cyber Engineering, Xidian University · Hangzhou Institute of Technology, Xidian University · Department of Computer Science, Purdue University

Abstract

Trust prediction provides valuable support for decision-making, risk mitigation, and system security enhancement. Recently, Graph Neural Networks (GNNs) have emerged as a promising approach for trust prediction, owing to their ability to learn expressive node representations that capture intricate trust relationships within a network. However, current GNN-based trust prediction models face several limitations: (i) Most of them fail to capture trust dynamicity, leading to questionable inferences. (ii) They rarely consider the heterogeneous nature of real-world networks, resulting in a loss of rich semantics. (iii) None of them support context-awareness, a basic property of trust, making prediction results coarse-grained. To this end, we propose CAT, the first Context-Aware GNN-based Trust prediction model that supports trust dynamicity and accurately represents real-world heterogeneity. CAT consists of a graph construction layer, an embedding layer, a heterogeneous attention layer, and a prediction layer. It handles dynamic graphs using continuous-time representations and captures temporal information through a time encoding function. To model graph heterogeneity and leverage semantic information, CAT employs a dual attention mechanism that identifies the importance of different node types and nodes within each type. For context-awareness, we introduce a new notion of meta-paths to extract contextual features. By constructing context embeddings and integrating a context-aware aggregator, CAT can predict both context-aware trust and overall trust. Extensive experiments on three real-world datasets demonstrate that CAT outperforms five groups of baselines in trust prediction, while exhibiting strong scalability to large-scale graphs and robustness against both trust-oriented and GNN-oriented attacks.

Figures & tables

Models	Guardian [ 6 ]	Medley [ 7 ]	GATrust [ 8 ]	TrustGNN [ 9 ]	KGTrust [ 10 ]	DTrust [ 11 ]	TrustGuard [ 2 ]	CAT
Dynamicity	○	●	○	○	○	●	●	●
Heterogeneity	○	○	○	○	●	○	○	●
Context-Awareness	○	○	○	○	○	○	○	●
Robustness	○	○	○	○	○	○	◐	●

TABLE I: Comparison between CAT and existing GNN-based trust prediction models.

Fig. 1: A motivating example of dynamic heterogeneous networks represented by a real-world social network with interactions involving users $u_{1}\sim u_{4}$ and items across various contexts $c_{1}\sim c_{3}$ , including sports, electronics, and clothing. Dashed and solid arrows indicate interactions occurring at different time points.

Datasets	# Users	# Items	# Ratings	# Trust relationships	# Contexts	Timestamps of ratings	Timestamps of trust relationships
Epinions	9163	12573	265189	311158	25	✓	✓
Ciao	2378	16861	36065	57544	6	✓	✗
CiaoDVD	19533	16121	72665	40133	17	✓	✗

TABLE II: Statistics of Epinions, Ciao, and CiaoDVD datasets.

Fig. 3: Results of context-aware trust prediction on the Ciao dataset. Case 1 illustrates a user pair $\langle v_{i},v_{j}\rangle$ with an overall distrusted relationship, whereas case 2 depicts a user pair with an overall trusted relationship.

CAT and variants	Epinions: Observed users			Epinions: Unobserved users			Ciao
CAT and variants	MRR	AP	AUC	MRR	AP	AUC	MRR	AP	AUC
CAT	0.6025	0.9383	0.9677	0.4082	0.9527	0.8933	0.4150	0.9234	0.9327
w/o Time Embedding	0.5575	0.9240	0.9597	0.2441	0.9040	0.8045	0.3702	0.9050	0.9142
w/o Type Attention	0.5941	0.9368	0.9677	0.3033	0.9393	0.8776	0.4035	0.9216	0.9322
w/o Node Attention	0.5781	0.9328	0.9653	0.3975	0.9501	0.8851	0.4042	0.9194	0.9293
w/o Ca Meta-path	0.5763	0.9274	0.9600	0.2924	0.9328	0.8592	0.3959	0.9157	0.9256

TABLE V: Ablation studies on CAT.

CAT and variants	Epinions: Observed users			Epinions: Unobserved users			Ciao
CAT and variants	MRR	AP	AUC	MRR	AP	AUC	MRR	AP	AUC
CAT	0.6025	0.9383	0.9677	0.4082	0.9527	0.8933	0.4150	0.9234	0.9327
w/o Time Embedding	0.5575	0.9240	0.9597	0.2441	0.9040	0.8045	0.3702	0.9050	0.9142
w/o Type Attention	0.5941	0.9368	0.9677	0.3033	0.9393	0.8776	0.4035	0.9216	0.9322
w/o Node Attention	0.5781	0.9328	0.9653	0.3975	0.9501	0.8851	0.4042	0.9194	0.9293
w/o Ca Meta-path	0.5763	0.9274	0.9600	0.2924	0.9328	0.8592	0.3959	0.9157	0.9256

TABLE V: Ablation studies on CAT.

Tasks	Models	Clean	Trust-oriented Attacks					GNN-oriented Attacks
Tasks	Models	Clean	$p$ =5%	$p$ =10%	$p$ =15%	$p$ =20%	MDR $\downarrow$	$p$ =5%	$p$ =10%	$p$ =15%	$p$ =20%	MDR $\downarrow$
①	Medley	0.4762	0.4552	0.4367	0.4155	0.4079	14.34%	0.4650	0.4576	0.4477	0.4393	7.75%
	TrustGuard	0.4955	0.4831	0.4820	0.4565	0.4528	8.62%	0.4813	0.4711	0.4664	0.4721	5.87%
	CAT	0.6025	0.5968	0.6046	0.6144	0.6070	0.95%	0.5999	0.5869	0.5842	0.5821	3.39%
②	Medley	0.1979	0.1894	0.1791	0.1756	0.1715	13.34%	0.1909	0.1890	0.1846	0.1737	12.23%
	TrustGuard	0.2571	0.2465	0.2431	0.2253	0.2250	12.49%	0.2477	0.2296	0.2348	0.2289	10.97%

TABLE VI: Robustness comparison under trust-oriented and GNN-oriented attacks (MRR results).

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Metrics	Tasks	Models	Clean	Trust-oriented Attacks					GNN-oriented Attacks
Metrics	Tasks	Models	Clean	$p$ =5%	$p$ =10%	$p$ =15%	$p$ =20%	MDR $\downarrow$	$p$ =5%	$p$ =10%	$p$ =15%	$p$ =20%	MDR $\downarrow$
AP	①	Medley	0.8944	0.8862	0.8855	0.8728	0.8613	3.70%	0.8923	0.8838	0.8827	0.8810	1.50%
		TrustGuard	0.8919	0.8860	0.8855	0.8724	0.8685	2.62%	0.8853	0.8816	0.8826	0.8814	1.18%
		CAT	0.9383	0.9394	0.9391	0.9415	0.9401	-0.09%	0.9399	0.9350	0.9326	0.9335	0.61%
	②	Medley	0.8884	0.8714	0.8473	0.8430	0.8328	6.26%	0.8799	0.8758	0.8780	0.8652	2.61%
		TrustGuard	0.8950	0.8918	0.8918	0.8824	0.8802	1.65%	0.8901	0.8881	0.8898	0.8881	0.77%

Appendix

TABLE VII: Robustness comparison under trust-oriented and GNN-oriented attacks (AP and AUC results).

Fig. 9: Visualization of the dual attention scores, including type attention scores (shown in blue) and node attention scores (shown in red). $v_{0}\sim v_{3}$ denote users and $v_{4}\sim v_{6}$ denote items, with $v_{0}$ being the target node.

Explore similar work

TCHG: Tri-Trust Conditioned Heterogeneous Graph Learning for Reliable Dynamic Trust Prediction
Jun 15, 2026Bohao Liao, Boyu Deng, Qipeng Song +2Log-Ratio-Based Gaussian Trust Weight Text-Attributed Graph
School of Telecommunication Engineering, Xidian University, Xi’an, 710071, Shaanxi, China · National Key Laboratory of Multi-domain Data Collaborative Processing and Control, Beijing, 100141, China · School of Cyber Engineering, Xidian University, Xi’an, 710071, Shaanxi, China
A Hybrid TGN-SEAL Model for Dynamic Graph Link Prediction
Feb 15, 2026Nafiseh Sadat Sajadi, Behnam Bahrak, Mahdi Jafari SiavoshaniZero-Shot Link Prediction Temporal Graph Neural Networks
Department of Computer Engineering, Sharif University of Technology, Tehran, Iran. · Tehran Institute for Advanced Studies, Khatam University, Tehran, Iran.
Dynamic Link Prediction with Temporally Enhanced Signed Graph Neural Networks
May 25, 2026Derek Regier, Andrew Polyak, Aresh Dadlani +1Temporal Graph Neural Networks Zero-Shot Link Prediction

Jun 15, 2026, cs.LGJ/K move · Enter open · S save

TCHG: Tri-Trust Conditioned Heterogeneous Graph Learning for Reliable Dynamic Trust Prediction

Bohao Liao, Boyu Deng, Qipeng Song, Jieling Wang+1

School of Telecommunication Engineering, Xidian University, Xi’an, 710071, Shaanxi, China · National Key Laboratory of Multi-domain Data Collaborative Processing and Control, Beijing, 100141, China · School of Cyber Engineering, Xidian University, Xi’an, 710071, Shaanxi, China

Log-Ratio-Based Gaussian Trust Weight Text-Attributed Graph AI Trustworthiness Recommendation

Open paper AskPDF arXiv

Trust prediction infers latent user-user trust relations and provides important support for social recommendation, fake-review and manipulation detection, and risk identification. Graph neural networks have become a prominent approach to trust prediction because of their ability to learn network structures and complex trust dependencies. However, existing methods often rely on a unified representation of trust signals and do not disentangle heterogeneous trust evidence into separate evidence channels, failing to exploit the distinct roles that different evidence channels should play during trust modeling. To address this gap, this paper argues that trust evidence should not be treated as an undifferentiated input, but should be decomposed and used as functional control factors over graph propagation. We propose TCHG, a tri-trust conditioned heterogeneous graph learning framework that decomposes trust evidence into three channels and assigns them distinct functional roles in propagation: entity reliability governs message admission, interaction-behavior reliability modulates propagation strength, and contextual trust adjusts the propagation mode through context-conditioned operator selection. Since the three evidence channels evolve at different temporal scales, TCHG maintains independent temporal states with non-uniform decay rates to prevent rapidly changing contextual signals from overwriting slowly accumulated entity reliability. It further predicts trust probability and calibrates the output probability, improving predictive confidence under sparse or conflicting evidence. Extensive experiments on multiple public trust datasets show that TCHG achieves effective and reliable trust prediction compared with representative trust prediction and heterogeneous graph baselines.

Abstract

Explore similar work

TCHG: Tri-Trust Conditioned Heterogeneous Graph Learning for Reliable Dynamic Trust Prediction

A Hybrid TGN-SEAL Model for Dynamic Graph Link Prediction

Dynamic Link Prediction with Temporally Enhanced Signed Graph Neural Networks