27 February 2024 in Knowledge-Based Systems
23 November 2023 in Biomedical Signal Processing and Control
21 March 2024 in IEEE Intelligent Systems
Multimodal knowledge graphs (MKGs) organize multimodal facts in the form of entities and relations, and have been successfully applied to several downstream tasks. Since most MKGs are incomplete, the MKG completion (MKGC) task has been proposed to address this problem, which aims to complete missing entities in MKGs. Previous most works obtain reasoning ability by capturing the correlation between target triplets and related images, but they ignore contextual semantic information and the reasoning process is not easily explainable. To address these issues, we propose a novel text-enhanced transformer fusion network called TE-TFN, which converts the context path between head and tail entities into natural language text and fuses multimodal features from both coarse and fine granularities through a multi-granularity fuser. It not only effectively enhances text semantic information, but also improves the interpretability of the model by introducing paths. Experimental results on benchmark datasets demonstrate the effectiveness of our model.
24 May 2025 in Biomedical Signal Processing and Control
27 October 2025 in IEEE Transactions on Intelligent Transportation Systems
Traffic forecasting leverages multivariate time series analysis to predict traffic patterns. Real-world traffic data comprises two distinct types of latent time-series signals: diffuse signals, which refer to time-varying information propagated across the traffic network, and intrinsic signals, which capture unique, location-specific patterns. However, existing approaches often treat traffic signals solely as diffusion outcomes, overlooking the intrinsic characteristics that can significantly influence model performance. To address this issue, we propose the Position-invariant Graph Convolutional Recurrent Network (PGCRN), which decouples diffuse and intrinsic signals for improved traffic forecasting. Instead of relying on a predefined graph, PGCRN learns graph structures from spatio-temporal data through a learnable position-invariant node representation that forms an adaptive adjacency matrix. This is integrated into a Graph Convolutional Recurrent Network (GCRN) encoder–decoder to jointly capture spatial and temporal dependencies. Furthermore, we introduce a contrastive learning framework in which a node’s time-varying and position-invariant representations form positive pairs, while position-invariant representations from different nodes form negative pairs. The model is trained with a triplet loss. Experiments on four benchmark datasets show that PGCRN consistently outperforms strong baselines. Owing to its computational efficiency, PGCRN is also well suited for deployment on resource-constrained edge devices.
13 March 2026 in IEEE Transactions on Computational Social Systems
Recommendation systems aim to enhance user satisfaction by providing personalized recommendations based on historical behaviors. Sequential recommendation, capable of capturing temporal dynamics and predicting future actions, has gained significant attention. However, existing methods typically focus on individual sequences and overlook collaboration between sequences, which limits their effectiveness. To address this, we propose AMCL4SR, an adaptive multitask collaborative learning framework optimized via Pareto optimization (PO). The entire framework is optimized under the paradigm of multitask learning using PO. First, we adopt an adaptive approach to learn a global collaborative graph among all items and employ an unsupervised strategy to leverage this graph for capturing global collaborative information. Furthermore, we leverage dynamic positional encoding to capture temporal dependencies and global contextual information, optimizing the interaction between sequential and global information. Finally, the framework is optimized using the multitask learning paradigm, enabling the modules of AMCL4SR to work together and improve performance. Experimental results show that AMCL4SR significantly outperforms baseline methods, demonstrating strong generalizability, effective global information modeling, and mitigation of embedding dimensionality collapse.