Learning dynamic user preference has become an increasingly important component for many online platforms (e.g., video-sharing sites, e-commerce systems) to make sequential recommendations. Previous works have made efforts model item-item transitions over interaction sequences, based on various architectures, e.g., recurrent neural networks and self-attention mechanism. Recently emerged graph also serve as useful backbone models capture item dependencies in recommendation scenarios. Despite...

Heterogeneous graph convolutional networks have gained great popularity in tackling various network analytical tasks on heterogeneous data, ranging from link prediction to node classification. However, most existing works ignore the relation heterogeneity with multiplex between multi-typed nodes and different importance of relations meta-paths for embedding, which can hardly capture structure signals across relations. To tackle this challenge, work proposes a Multiplex Graph Convolutional...

Abstract Recommending a limited number of Point-of-Interests (POIs) user will visit next has become increasingly important to both users and POI holders for Location-Based Social Networks (LBSNs). However, recommendation is challenging task since complex sequential patterns rich contexts are contained in extremely sparse check-in data. Recent studies show that embedding techniques effectively incorporate contextual information alleviate the data sparsity issue, Recurrent Neural Network (RNN)...

Network embedding has emerged as a new learning paradigm to embed complex network into low-dimensional vector space while preserving node proximities in both structures and properties. It advances various mining tasks, ranging from link prediction classification. However, most existing works primarily focus on static networks many real-life evolve over time with addition/deletion of links nodes, naturally associated attribute evolution. In this work, we present Motif-preserving Temporal...

Heterogeneous graph neural networks have gained great popularity in tackling various network analysis tasks on heterogeneous data. However, most existing works mainly focus general networks, and assume that there is only one type of edge between two nodes, while ignoring the multiplex characteristics multi-typed nodes different importance structures among for node embedding. In addition, over-smoothing issue limits models to capturing local structure signals but hardly learning global...

Traffic prediction is critical for optimizing travel scheduling and enhancing public safety, yet the complex spatial temporal dynamics within traffic data present significant challenges accurate forecasting. In this paper, we introduce a novel model, Spatiotemporal-aware Trend-Seasonality Decomposition Network (STDN). This model begins by constructing dynamic graph structure to represent flow incorporates spatio-temporal embeddings jointly capture global dynamics. The representations learned...

Traffic forecasting is crucial for intelligent transportation systems (ITS), aiding in efficient resource allocation and effective traffic control. However, its effectiveness often relies heavily on abundant data, while many cities lack sufficient data due to limited device support, posing a significant challenge forecasting. Recognizing this challenge, we have made noteworthy observation: patterns exhibit similarities across diverse cities. Building key insight, propose solution the...

Traffic forecasting is a critical service in Intelligent Transportation Systems (ITS). Utilizing deep models to tackle this task relies heavily on data from traffic sensors or vehicle devices, while some cities might lack device support and thus have few available data. So, it necessary learn data-rich transfer the knowledge data-scarce order improve performance of forecasting. To address problem, we propose cross-city few-shot framework via Pattern Bank (TPB) due that patterns are similar...

Accurate citywide traffic inference is critical for improving intelligent transportation systems with smart city applications. However, this task very challenging given the limited training data, due to high cost of sensor installment and maintenance across entire urban space. A more practical scenario study effectively modeling spatial temporal patterns historical observations. In work, we propose a dynamic multi-view graph neural network method CTVI+. Specifically, dimension,...

Next location recommendation plays an important role in various location-based services, yielding great value for both users and service providers. Existing methods usually model temporal dependencies with explicit time intervals or learn representation from customized point of interest (POI) graphs rich context information to capture the sequential patterns among POIs. However, this problem is perceptibly complex, because factors, e.g., users’ preferences, spatial locations, contexts,...

Inferring social relationships from user location data has become increasingly important for real-world applications, such as recommendation, advertisement targeting, and transportation scheduling. Most existing mobility relationship measures are based on pairwise meeting frequency, that it, the more frequently two users meet (i.e., co-locate at same time), likely they friends. However, frequency-based methods suffer greatly sparsity challenge. Due to collection limitation bias in real world...

In recent years, heterogeneous network representation learning has attracted considerable attentions with the consideration of multiple node types. However, most them ignore rich set attributes (attributed network) and different types relations (multiplex network), which can hardly recognize multi-modal contextual signals across interactions. While a handful embedding techniques are developed for attributed multiplex networks, they significantly limited to scalability issue on large-scale...

Trajectory-User Linking (TUL), which links trajectories to users who generate them, has been a challenging problem due the sparsity in check-in mobility data. Existing methods ignore utilization of historical data or rich contextual features data, resulting poor performance for TUL task. In this paper, we propose novel Mutual distillation learning network solve sparse named MainTUL. Specifically, MainTUL is composed Recurrent Neural Network (RNN) trajectory encoder that models sequential...

Heterogeneous graph convolutional networks have gained great popularity in tackling various network analytical tasks on heterogeneous data, ranging from link prediction to node classification. However, most existing works ignore the relation heterogeneity with multiplex between multi-typed nodes and different importance of relations meta-paths for embedding, which can hardly capture structure signals across relations. To tackle this challenge, work proposes a M ultiplex H eterogeneous G raph...