Recommender systems are essential for filtering content to match user preferences. However, traditional recommender systems often suffer from biases inherent in the data, such as popularity bias. These biases, particularly those stemming from latent confounders, can result in inaccurate recommendations and reduce both the diversity and effectiveness of the system. Existing debiasing methods for recommender systems, however, either fail to account for latent confounders or rely on predefined instrumental variables (IVs). To address this research gap, we propose a novel causality-based recommendation algorithm, Data-driven IV representation learning for debiasing in Recommender System (DIVRS), which enables the learning of IV representation directly from user-item interaction data. By leveraging the learned IV representation, DIVRS decomposes user behaviour into causal and confounding relationships to address potential bias in recommender systems. Additionally, we introduce Orthogonal Promotion Regularisation (OPR) for DIVRS to address the problem that Graph Convolutional Networks (GCNs) amplify bias. We also propose a variant of GCNs for DIVRS, called DIVRS-GCN. Experimental results on the Douban-Movie and Movielens-10M datasets demonstrate that both DIVRS and DIVRS-GCN effectively mitigate confounding bias while outperform the state-of-the-art methods in recommendation performance. For example, on both datasets, our DIVRS and DIVRS-GCN improve Recall@20 by up to 10.98 %. This validates their effectiveness and robustness. Our approaches improve recommendation accuracy while delivering more balanced and diverse suggestions, effectively addressing the limitations of existing IV-based recommender systems.

Understanding vehicle-to-vehicle interactions is critical for safe and efficient autonomous driving in complex traffic scenarios. However, most existing behavior prediction and planning models implicitly account for interactions through soft weighting or metrics based on distance and risk, without a clear or quantitative definition of vehicle interaction. This paper proposes a metric called Vehicle Interaction Level (VIL) to quantify the degree to which a vehicle’s behavior is constrained by another vehicle. By applying a VIL threshold, strong-interaction vehicles are distinguished from weak ones. Then, the Feature Difference Level (FDL) is proposed to analyze feature-level behavioral differences across interaction categories, which is quantified using the Random Forest (RF) and the Gini Index (GI). Based on real-world vehicle trajectory pairs, an optimal VIL threshold of 0.2 is determined by maximizing FDL to identify strong-interaction vehicles, providing a more discriminative and broadly applicable interaction characterization than distance-based and risk-based metrics. Furthermore, the application of this strong-interaction vehicle identification method to trajectory prediction achieves consistently better performance than the baseline across multiple evaluation metrics, demonstrating its practical value for downstream autonomous driving tasks. Owing to its lightweight and model-agnostic formulation, the VIL-based method can be integrated into existing autonomous driving systems operating in highly interactive traffic scenarios.

Graph Neural Networks (GNNs) have made significant progress in graph mining tasks, such as node classification. In particular, graph contrastive learning (GCL) methods based on data augmentation have been proven to effectively improve model performance. Existing GCL typically use two technologies: feature augmentation and structure augmentation. However, these technologies often cause excessive perturbation in the feature space when transforming node features. In addition, many structure augmentation techniques disrupt the graph topology, and many methods ignore the multi-hop neighbor information of nodes. To address these issues, we propose a new method called Spectral Feature Augmentation-based Multi-Hop Contrastive Learning (SMHCL), for node classification. First, we use spectral feature augmentation to implicitly inject noise into the singular values of node features, which avoids excessive perturbation in the feature space. Next, we generate multiple views by aggregating the multi-hop neighbor information of nodes. This approach effectively captures extensive neighbor information while preserving the graph’s topology. Finally, we design a multi-hop contrastive loss function based on graph contrastive learning across multiple graphs. This enhances representation learning for the node classification task. Extensive experimental results show that SMHCL significantly outperforms existing GCL methods, both in label-rich and label-scarce scenarios.

It is often challenging to identify a valid instrumental variable (IV), although the IV methods have been regarded as effective tools of addressing the confounding bias introduced by latent variables. To deal with this issue, an Interaction-Data-guided Conditional IV (IDCIV) debiasing method is proposed for Recommender Systems, called IDCIV-RS. The IDCIV-RS automatically generates the representations of valid CIVs and their corresponding conditioning sets directly from interaction data, significantly reducing the complexity of IV selection while effectively mitigating the confounding bias caused by latent variables in recommender systems. Specifically, the IDCIV-RS leverages a variational autoencoder (VAE) to learn both the CIV representations and their conditioning sets from interaction data, followed by the application of least squares to derive causal representations for click prediction. Extensive experiments on two real-world datasets, Movielens-10M and Douban-Movie, demonstrate that IDCIV-RS successfully learns the representations of valid CIVs, effectively reduces bias, and consequently improves recommendation accuracy.

Graph similarity computation plays a crucial role in a variety of fields such as chemical molecular structure comparison, social network analysis and code clone detection. However, due to inadequate feature representation, existing methods often struggle to cope with complex graph structures, which in turn limits the feature fusion capability and leads to low accuracy of similarity computation. To address these issues, this paper introduces an Adaptive Multi-scale Feature Fusion(AMFF) framework. AMFF firstly enhances feature extraction through a residual graph neural network, which robustly captures key information in complex graph structures. Based on this, a multi-pooled attention network is used to aggregate multi-scale features and accurately extract key node features while minimizing information loss. Finally, the adaptive multi-scale feature fusion mechanism dynamically adjusts the feature fusion weights according to the interactions between nodes and graph embeddings, thus improving the accuracy and sensitivity of similarity computation. Extensive experiments on benchmark datasets including AIDS700nef, LINUX, IMDBMulti, and PTC show that AMFF significantly outperforms existing methods on several metrics. These results confirm the efficiency and robustness of AMFF in graph similarity computation, providing a promising solution for assessing the similarity of complex graph data.

The Automatic Emergency Braking (AEB) system is capable of assessing driving risks, alerting the driver to potential collision hazards, and, in the absence of driver response to the collision risk, autonomously activating braking to mitigate the occurrence of collision accidents. Most existing Automatic Emergency Braking (AEB) systems rely on Time to Collision (TTC) for risk assessment and decision-making. However, TTC fails to account for the impact of absolute velocity on driving safety when assessing risk, leading to inaccurate risk descriptions, particularly in high-speed scenarios with minor speed differences. The Safety Margin (SM) takes into account key factors affecting driving risk, such as relative velocity and distance, and is capable of accurately quantifying driving risks. Based on the SM, this study proposes a full-speed range single-threshold Automatic Emergency Braking (AEB) model. The model comprises two components: traffic environment risk quantification and road surface friction coefficient estimation. It is applicable to automatic emergency braking tasks under varying speeds and road surface conditions. Simulation experiments were conducted by constructing three typical scenarios: stationary lead vehicle, slow-moving lead vehicle, and braking lead vehicle, to determine the braking threshold as 0.2. The safety performance of the proposed safety margin-based AEB model is evaluated by comparing it with the traditional TTC-based AEB model across the specified scenarios. The results demonstrate that the safety margin-based AEB model proposed in this study achieves 100% safe braking in all scenarios, successfully performing emergency braking and outperforming the TTC-based AEB model.

Hyper-relational knowledge graphs (HKGs) extend the traditional triplet-based knowledge graph by adding qualifiers to the relationships, making HKGs particularly useful for tasks that require more profound understanding and inference from relationships between entities. However, existing hyper-relational knowledge representation learning methods (HKRL) focus on direct neighbourhood information of entities only by neglecting the relational similarity of the main triple in hyper-relational facts and the attribute details in the qualifiers. In addition, few works extract common and private information across multiple views to minimize noise and interference. This paper proposes a multi-hypergraph disentanglement method for HKRL to address the above issues. Specifically, we first construct four hypergraphs to mine and utilise the inherent structure information of HKGs, and then propose to extract common representations among hypergraphs and private representations within individual hypergraphs to mine the semantic information and the task-relevant information, respectively. Experiment results on four real datasets demonstrate the effectiveness of the proposed method compared to SOTA methods in link prediction tasks on HKGs.

In deep contrastive graph clustering, many methods tend to adopt a singular enhancement strategy, focusing either on structural or attribute augmentation. This limited approach constrains the model's ability to integrate multidimensional information, resulting in an imbalance in information utilization. Furthermore, the randomness involved in the selection of negative samples may lead to blurred distinctions between positive and negative samples. Therefore, we introduce a graph attention-based dual enhancement multiview clustering (GA-DE-MVC). The GA-DE-MVC first encodes structure and attributes through attention mechanisms and fully connected layers to achieve dual enhancement of structure and attributes. Then, it selects the farthest sample as the negative sample based on the Euclidean distance between cluster centers, to alleviate the randomness in negative sample selection, thereby enhancing the distinctiveness between positive and negative samples.The experimental results on six datasets surpass existing algorithms, verifying the effectiveness of the GA-DE-MVC algorithm.

In most studies on modeling driving behavior at uncontrolled intersections, multi-vehicle interaction scenarios are usually categorized and modeled separately as moving-across behavior and merging behavior. However, it is inappropriate to use a single-behavior model to accurately represent general driving behavior in uncontrolled intersections. In this case, we constructed a general driving behavior model for multi-vehicle interaction at uncontrolled intersections. Initially, the IMM model is employed to anticipate the movement of the vehicle within the driver’s visual field. The risk field theory is applied to assess potential hazards that the vehicle might confront, drawing from the risk homeostasis theory and preview-follower theory, which aids in determining a trajectory that aligns with the drivers’ real-life actions while also meeting the risk constraints. Drivers’ heterogeneity is reflected by risk threshold. This model can simulate driver behavior in traffic congestion at uncontrolled intersections by adjusting risk thresholds when the vehicles are caught in a deadlock situation. Results show that our model can accurately reproduce the priority and trajectory of vehicles crossing the intersection and resolve multi-vehicle conflicts within a reasonable time. This model can be used for traffic simulation at uncontrolled intersections and to provide test validation for automated driving systems.

Graph neural networks (GNNs) have achieved promising results for semi-supervised learning tasks on the graph-structured data. However, most existing methods assume that the training data are with correct labels, but in the real world, the graph-structured data often carry noisy labels to reduce the effectiveness of GNNs. To address this issue, this paper proposes a new label correction method, called multi-teacher self-training (MTS-GNN for short), to conduct semi-supervised node classification with noisy labels. Specifically, we first save the parameters of the model training in the earlier iterations as teacher models, and then use them to guide the processes, including model training, noisy label removal, and pseudo-label selection, in the later iterations of the training process of semi-supervised node classification. As a result, based on the guidance of the teacher models, the proposed method achieves the model effectiveness by solving the over-fitting issue, improves the accuracy of noisy label removal and the quality of pseudo-label selection. Extensive experimental results on real datasets show that our method achieves the best effectiveness, compared to state-of-the-art methods.

There is a growing number of studies on the traffic control strategies of signal timings and vehicle trajectories at signalized intersections, while lane assignments are widely pre-specified and fixed. Meanwhile, existing strategies generally require a fully connected and automated vehicles (CAVs) environment. To fill up the gaps, this study contributes to a two-dimensional (spatiotemporal) control strategy by jointly optimizing traffic signals, lane settings, and vehicle trajectories at isolated signalized intersections under the mixed traffic of connected automated and human-driven vehicles. Specifically, based on the pseudo-platoons, signal timing plans and settings of approach lanes are jointly optimized by a piece-wise linear programming model. Then, vehicle trajectory control is integrated into the collaborative control framework to smooth vehicle trajectories. Three groups of numerical experiments are conducted to verify the effectiveness and efficiency of the proposed control method. Results show that the proposed algorithm outperforms the actuated control in terms of vehicle travel time under both under-saturated and over-saturated traffic conditions.

Although it’s very common nowadays to experience onboard adaptive cruise control systems (ACCS), most of them are initialized with fixed parameters and little consideration is taken to adapting to different drivers. In order to improve the ride comfort and operation familiarity of drivers when using the ACCS, driver’s personalized car-following (CF) characteristics should be considered in ACCS developments. In this paper, an ACCS framework with CF characteristics identification and application is designed, where the desired safety margin (DSM) model is utilized to describe driver’s CF behaviors. In view of the limited performance of on-board computers, a simple method for CF characteristics identification from the sight of driver’s physiology and psychology mechanism is proposed, which tries to estimate DSM model parameters on the basis of CF data. With the proposed method, the CF data under manual driving are collected and filtered for the identification of drivers’ response time, upper and lower limits of DSM and sensitivity factors for acceleration and deceleration iteratively. And the latest iteration results of DSM model parameters would then be applied into the ACCS when it’s enabled. Several naturalistic driving tests are adopted for the verification whose results support that, the proposed method could perform well in identifying drivers’ CF characteristics online by collecting and analyzing manual driving data with a much smaller resource consumption when compared with genetic algorithm (GA). Besides, with the increase of CF data volume, the identified results are also becoming more and more stable and robust. Therefore, the proposed ACCS framework is able to imitate different drivers.

Since vehicles are not restricted by lane lines when driving inside intersections, they travel freely and their paths have many possibilities, which are related to the traffic order of other road users. It is necessary to understand and describe vehicle paths through intersections. Previous studies on modeling vehicle paths through intersections have predominantly been limited to vehicles turning in one single direction or specific intersections, ignoring the application universality. This study set out to establish a unified model to describe non-conflict paths of vehicles through intersections. An entrance lane-based coordinate system with geometric parameters was established to describe the position relationships of entrance and exit lanes. Three-order Bezier curves with control parameters were used to describe vehicle paths. By extracting the parameter values from driving simulator-based experiments, the entrance and exit lane-based vehicle path model was proposed to represent the relationship between geometric and control parameters. Actual path data captured by an unmanned aerial vehicle were used for model validation, and results show the validity of the proposed model with a mean Root Mean Square Error of 0.4613 m between observed and fitted paths. More validation considering position distributions of path origins and destinations and angle deviations when entering and leaving intersections was performed. Cases from HighD open dataset also demonstrated the model’s potential for lane-changing scenarios. This study provides an exciting opportunity to commonly describe vehicle paths without conflicts at geometrically regular and irregular intersections and is hoped to have the possibility of planning local paths for autonomous vehicles.

Unsupervised graph representation learning based on graph autoencoder and graph variational autoencoder has achieved significant success in non-Euclidean data such as citation networks, social networks, and so on. However, the most existing graph autoencoders aggregate node features and graph structure from one view: local topology, and only reconstruct the node feature matrix or adjacency matrix, which neither learns a more useful and comprehensive embedding nor makes full use of the latent information of the embedding. In this paper, we propose a multi-view graph autoencoder which can aggregate latent information from local topology, global topology and feature similarity and reconstruct the graph structure and node features simultaneously. We validate the effectiveness of our framework on four datasets and the experimental results demonstrate the superior performance of our proposed framework compared with other advanced frameworks.

Microscopic modeling of driving behavior is the basis for traffic design and traffic simulation studies and can be applied to automated driving systems to provide human-like decision making. Previous modeling methods can be mainly divided into scenario-based modeling methods and field theory-based modeling methods. Scenario-based models are based on behavior theories that can explain behavioral mechanisms and field theory-based models are convenient for application to different scenarios. Combining two behavior theories and field theory, this paper aims to present a novel method to uniformly model the driving behavior in different scenarios. Risk homeostasis theory and preview-follower theory are used as the theoretical foundation, and field theory is utilized to connect the two behavior theories. A new risk field model is constructed for better coupling these behavior theories. Integrating these theories, this study then develops a subjectively perceived risk quantification method and a trajectory and motion planning model, which are validated using naturalistic data in car-following scenarios. Results show the effectiveness of this method and this model with reference to an effective risk quantification index (safety margin) and in comparison with the classical models (desired safety margin model and intelligent driver model) using naturalistic data in car-following scenarios.

In recent years, the adaptive cruise control (ACC) system has become widely adopted. The vehicle-positioning system plays an important role in automotive platoon driving and can provide vehicle-position information through vehicle-to-vehicle communication. However, the acquired vehicle-position information often presents uncertainties given the existence of variations in the traffic environment and of noise in the vehicle-positioning system. These uncertainties, in turn, influence the car-following performance of a vehicle platoon. In this study, we employ the desired safety margin (DSM) model, an ACC control strategy, to investigate the influence of the uncertainty level of vehicle-position information (ULVPI) on string stability and car-following safety. The stability criterion of the DSM model with ULVPI is derived through linear stability theory. Theoretical analysis results are verified by numerical simulations. Analytical results indicate that a negative ULVPI value can expand the stable region and improve string stability and that a positive ULVPI value can reduce delay time. Moreover, a negative ULVPI value can improve car-following safety during the stopping and evolution processes, whereas a positive ULVPI values can increase the safety margins of vehicles during the starting process. In the car-following process, a negative ULVPI value can improve car-following safety and reduce rear-end collision risk. The variation in ULVPI values can improve string stability and reduce the risk of rear-end collisions when mean and standard deviation are reasonable. Therefore, the improvement of a vehicle’s dynamic performance, string stability, and safety relies on the effect of ULVPI in different traffic scenarios. These results are useful in designing different control strategies that stabilize traffic flow and improve traffic safety for vehicles with ACC systems.

Abstract Aspect sentiment classification is an important research topic in natural language processing and computational linguistics, assisting in automatically review analysis and emotional tendency judgement. Different from extant methods that focus on text sequence representations, this paper presents a network framework to learn representation from concurrence-words relation graph (LRCWG), so as to improve the Macro-F1 and accuracy. The LRCWG first employs the multi-head attention mechanism to capture the sentiment representation from the sentences which can learn the importance of text sequence representation. And then, it leverages the priori sentiment dictionary information to construct the concurrence relations of sentiment words with Graph Convolution Network (GCN). This assists in that the learnt context representation can keep both the semantics integrity and the features of sentiment concurrence-words relations. The designed algorithm is experimentally evaluated with all the five benchmark datasets and demonstrated that the proposed aspect sentiment classification can significantly improve the prediction performance of learning task.

It has been demonstrated that the segmentation performance is highly dependent on both subspace preservation and graph connectivity. In the literature, the full connectivity method linearly represents each data point ( e.g., a pixel in one image) by all data points for achieving subspace preservation, while the sparse connectivity method was designed to linearly represent each data point by a set of data points for achieving graph connectivity. However, previous methods only focused on either subspace preservation or graph connectivity. In this article, we propose a Sparse Graph Connectivity (SGC) method for image segmentation to automatically learn the affinity matrix from the low-dimensional space of original data, which aims at simultaneously achieving subspace preservation and graph connectivity. To do this, the proposed SGC simultaneously learns a self-representation affinity matrix for subspace preservation and a sparse affinity matrix for graph connectivity, from the intrinsic low-dimensional feature space of high-dimensional original data. Meanwhile, the self-representation affinity matrix is pushed to be similar to the sparse affinity as well as be the final segmentation results. Experimental result on synthetic and real-image datasets showed that our SGC method achieved the best segmentation performance, compared to state-of-the-art segmentation methods.

The application of short-term traffic forecasting can guide the operation of traffic networks efficiently and reduce the traffic cost for travelers. On the basis of radial basis function (RBF) neural network, this paper introduces a tunable and transferable RBF (TT-RBF) model to conduct on-line forecasting and transfer forecasting. Considering the spatiotemporal correlation of traffic flows in a road network, a spatiotemporal state matrix formed by the detrended cross-correlation analysis is used for the model input. With the on-line forecasting process, an improved on-line structure and parameter adjustment are proposed to enhance the existing model. Thus, the TT-RBF model can be adaptive to time-varying traffic states, especially to deal with the difference between non-peak and peak hours. Moreover, the proposed model can be transferred from one road segment to act on other road segments. By this way, the traffic states of numerous road segments can be forecasted conveniently without complex model training processes. The floating car data of a typical road network in Beijing are used for the performance verification of the TT-RBF model, and some frequently used forecasting models are selected for comparisons. The numerical experiments show that the TT-RBF model can get more accurate results than those in single-step forecasting, multi-step forecasting, and transfer forecasting.

Car following is the most common phenomenon in single-lane traffic. However, the propagation of the small perturbation of the velocity of the leading car will affect traffic flow. Driving behaviors play an important role in the determination of the qualitative dynamics of vehicles in the car-following process. Under different traffic environments, driving behaviors depend on the level of perceived risk, acceleration and deceleration habits, and reaction characteristics of the driver. The desired safety margin (DSM) model can directly describe the driving behaviors in the car-following process by using the parameters of the risk perception of drivers, sensitivity coefficient of acceleration and deceleration, and response time. In this paper, we investigate the influence of the accepted risk level, response time, and sensitivity factor on the traffic flow via the DSM model. The stability criterion of the simplified DSM model is derived via linear stability theory. Analytical results indicate that a backward propagating of perturbation would enlarge or shrink with the change of three driving behavior parameters of accepted risk level, sensitivity coefficient of acceleration or deceleration, and response time. Results show that careful driving can improve the stability of traffic flow and that system stability can be maintained by adjusting the acceleration and deceleration control parameters, increasing DSM, or decreasing response time for the adaptive cruise control or vehicular platoon control system. The results can provide reasonable values of driving behavior parameters for the stability of the primitive DSM model using the simplified DSM model. Furthermore, we analyze the influence of interval DSMs, and the acceleration and deceleration sensitivity of the primitive DSM model on the stability of traffic flow through numerical simulations. Results imply that the lower limit of the DSM influences traffic flow more significantly than the upper limit of the DSM. Moreover, the increase in deceleration sensitivity has a more important influence on the stability of traffic flow than the increase in acceleration sensitivity. The numerical simulation results are in good agreement with the analytical results and the relevant experimental results of previous studies.