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

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.

💡 Research Summary

This paper proposes CAT, a novel Context-Aware GNN-based Trust prediction model designed to overcome the limitations of existing approaches in dynamic and heterogeneous networks. The core motivation stems from three identified gaps in current research: the failure to capture the dynamic evolution of trust over time, the oversight of rich semantic information in heterogeneous networks containing multiple node and edge types (e.g., users, items, trust relations, ratings), and the complete lack of support for context-awareness—a fundamental property where trust is specific to situations like item categories or interaction domains.

CAT features a layered architecture comprising a graph construction layer, an embedding layer, a heterogeneous attention layer, and a prediction layer. To model fine-grained trust dynamicity efficiently, it employs continuous-time representations through a time encoding function, coupled with a recent-time neighbor sampling strategy to ensure scalability. To handle network heterogeneity, it introduces a dual attention mechanism that learns the importance of different node types and the importance of individual nodes within each type, effectively filtering noise and leveraging semantic information. Most innovatively, to incorporate context-awareness, the authors propose the novel concept of “context-aware meta-paths” to integrate contextual features into the graph structure. They construct context embeddings by aggregating item embeddings within the same category and design a context-aware aggregator. This allows CAT to predict both fine-grained, context-specific trust scores and overall trust, even in the absence of labeled context-aware trust data, by learning the relationship between the two.

Extensive evaluations on three real-world datasets (Epinions, Ciao, CiaoDVD) demonstrate CAT’s superior performance. It outperforms five groups of baseline models, including static/dynamic and homogeneous/heterogeneous GNNs. The improvement is particularly significant for predicting trust for unobserved users (new users with no interactions during training), where CAT achieves a 50.79% improvement in Mean Reciprocal Rank (MRR) over the best baseline on the Epinions dataset. Furthermore, CAT exhibits strong scalability, reducing average running time by 73.97% on large graphs, and demonstrates remarkable robustness against both trust-oriented node-level attacks and more sophisticated GNN-oriented structural attacks, with maximum performance drops of only 0.95% and 3.39%, respectively.

In summary, CAT represents a significant advancement by being the first trust prediction model to jointly address dynamism, heterogeneity, and context-awareness using GNNs, backed by compelling empirical evidence of its effectiveness, scalability, and robustness.