Pedestrian trajectory prediction plays a vital role in autonomous driving and intelligent surveillance systems. Graph neural networks (GNNs) have shown remarkable effectiveness in this task by explicitly modeling social interactions among pedestrians. However, existing methods suffer from two key limitations. First, they face difficulty in balancing the reduction in redundant connections with the preservation of critical interaction relationships in spatial graph construction. Second, higher-order graph convolution methods lack adaptability to varying crowd densities. To address these limitations, we propose a pedestrian trajectory prediction method based on Delaunay triangulation and density-adaptive higher-order graph convolution. First, we leverage Delaunay triangulation to construct a sparse, geometrically principled adjacency structure for spatial interaction graphs, which effectively eliminates redundant connections while preserving essential proximity relationships. Second, we design a density-adaptive order selection mechanism that dynamically adjusts the graph convolution order according to pedestrian density. Experiments on the ETH/UCY datasets show that our method achieves 5.6% and 9.4% reductions in average displacement error (ADE) and final displacement error (FDE), respectively, compared with the recent graph convolution-based method DSTIGCN, demonstrating the effectiveness of the proposed approach.

Recently, the keyword-aware routing problem has been widely studied. The existing research primarily focuses on finding the Top-k optimal routes that satisfy the given origin and destination while covering the required query keywords through POIs (Point of Interest). However, since the routes are constructed solely based on distance, the POIs visited along these paths are highly similar. This fails to offer diverse routes and worsens traffic congestion. Therefore, in this paper, we study the Diversified Top-k Optimal Routes with Collective Spatial Keywords (Dk-ORCSK) problem, which finds k paths with similarity below a threshold τ and minimal total length. To avoid repeatedly visiting the same POIs during path expansion in existing methods, we propose a novel Region Deviation Algorithm (RDA). First, we pre-partition the road network into multiple regions. Next, starting from the origin, we satisfy the query keywords by deviating candidate paths toward the optimal route in different regions, rather than blindly expanding to individual POIs. However, blindly expanding to different regions still leads to higher path similarity. Therefore, we further analyze the similarity among candidate paths and introduce two pruning strategies to improve efficiency. Extensive experiments on real-world datasets demonstrate that our algorithm performs better in both efficiency and result quality.

The rapid advancement of artificial intelligence and the proliferation of online education platforms have made accurate and personalized recommendation services a critical component of modern learning systems. Traditional recommendation models often exhibit low accuracy due to the cold-start problem and the limited availability of user–item interaction data. Existing solutions have introduced knowledge graphs (KGs) to incorporate semantic relations into recommendation models. However, the improvement in accuracy remains limited. Moreover, KG-based methods commonly rely on locally constructed graphs that are incomplete and unable to exploit information distributed across other educational platforms. In this work, we propose PLRF, a federated knowledge-graph-based recommendation framework that enables collaborative learning across multiple platforms while preserving data privacy. To handle data heterogeneity, PLRF introduces a regularization mechanism during local training to align heterogeneous representations. Extensive evaluations on two real-world datasets demonstrate that PLRF achieves improvements in NDCG, Recall, Precision, and F1 score compared with both non-KG and single-source KG approaches, highlighting the potential of federated knowledge graph learning in cross-platform personalized recommendation.

Representation learning is key in trajectory analysis, affecting downstream task performance. To reduce computational overhead from numerous trajectory points, spatial partitioning of the road network is often used. Existing spatial partitioning methods primarily include grid-based partitioning and road network partitioning, which aim to achieve a more uniform distribution of trajectory points. However, after partitioning, these methods typically employ one-hot encoding to represent subregions, failing to capture the spatial proximity relationships between adjacent subregions. Moreover, due to variations in road network structures, existing trajectory representation learning methods often require complete retraining when adapting to new maps, resulting in high computational costs. This paper proposes a Partition Transition Probability and Pretraining Graph Attention Network (PTP-PGAT). The model integrates a road network partitioning strategy and incorporates a partition transition probability mechanism to capture spatial proximity between sub-regions effectively. Additionally, leveraging the hierarchical partition tree structure of road network partitioning, PTP-PGAT supports a pretraining-finetuning transfer learning paradigm. After pretraining on one dataset, the model can efficiently adapt to new datasets with minimal fine-tuning. Experiments show PTP-PGAT outperforms baselines, improving accuracy by up to 15.5

Trajectory-User Linking (TUL) task aims to accurately match anonymous trajectories to their corresponding users. As a critical task in mobility data mining, its resolution is crucial for a wide range of downstream applications, including personalized recommendations, urban planning, and public safety. However, existing methods primarily focus on point-level data while neglecting the holistic travel semantics embedded within urban road network. Moreover, they overlook the higher-order relationships among trajectories of different users. Consequently, we propose a Multiscale Semantics-Relationships Fusion Representation Model for TUL, namely MSRTUL. Specifically, we first propose a multiscale destination-oriented trajectory semantics encoder, which captures destination-oriented spatial-temporal semantics from both road-level and zone-level trajectories to encode the holistic travel semantics. Subsequently, we design a multiscale trajectory high-order relationships encoder, which jointly models higher-order relationships among trajectories, points, and categories through the hypergraph. To effectively combine the travel semantics and higher-order relationships, we design a dual-view fusion layer. We conduct extensive experiments on three real-world datasets, demonstrating that MSRTUL achieves significant improvements over multiple baselines in the TUL.

With the development of GPS positioning technology, location-based services, and intelligent transportation systems, keyword-aware optimal route (KAOR) planning has been widely studied. However, most existing studies primarily focus on returning the shortest route that satisfies keywords, which limits users’ ability to make personalized route selections. To address this issue, we propose UniRoute, a pioneering unified framework for KAOR that simultaneously supports top-k route planning, skyline route planning considering POI (Point of Interest) parking hardness, and diversity route planning considering POI similarity. To efficiently solve these problems, we first employ the G-Tree graph partitioning index to rapidly filter POIs and reduce the search space. We then propose a vertex deviation-aware splicing-based expansion algorithm, D-SBE, integrated with preprocessing techniques to improve efficiency. The generated candidate paths provide an accurate basis for different route queries. Finally, we design several optimization strategies tailored to the characteristics of the three query types to further improve performance. Extensive experiments on real-world road networks show that UniRoute achieves query times 1-2 orders of magnitude faster than existing methods while reducing route distances by an average of 20

Given a start point s, a destination t, and the set of keyword types that must be satisfied, the goal of Top-k Keywords-Aware Optimal Route (k-KAOR) query is to return top-k shortest routes that covers all query keyword types. Existing solutions can be divided into two classes: candidate based algorithms and path expansion based algorithms. However, they all face a trade-off between query efficiency and route quality. To solve this problem, in this paper, we further study the k-KAOR problem and introduce a Splice-Based Expansion (SBE) algorithm. The SBE algorithm efficiently integrates the ideas of candidate set and path expansion. First, to avoid a vast search space, we introduce a Point of Interest (POI) filtering refinement strategy based on the spatial index IG-Tree. Then, to cover more combinations of POIs while avoiding brute-force enumeration, we pre-construct a keyword distance matrix and splice POIs using the graph partitioning of the IG-Tree. Finally, we perform path expansion by concatenating different POI combinations until the top-k paths are discovered. Extensive experiments on real-world road networks show that our method can reduces response time by an order of magnitude and improves result quality by 25

Blockchain, valued for its decentralized and tamper-proof features, has sprung up as a predominant paradigm in numerous security applications. To facilitate massive trustworthy retrieval services, blockchain-assisted clouds, combining immutable on-chain storage and scalable off-chain storage, offer a data-reliable and storage-flexible platform. Based on the blockchain-assisted cloud model, existing studies have focused on several types of authenticated retrievals including verifiable skyline queries, graph queries, range queries, e.t.c., however, currently, few of them can directly support efficient query processing and result verification for trajectory similarity retrieval. To the best of our knowledge, this paper is the first attempt to address the problem of authenticated trajectory similarity retrieval on the blockchain-assisted cloud. To this end, we first propose novel authenticated data structures, 𝒫ℳℬ -Tree, to enable authenticated trajectory similarity retrieval, and 𝒮𝒫ℳℬ -Tree, to save gas consumption on the blockchain. To reduce the client-side verification costs, we further optimized the index as 𝒫ℳℬ^* -Tree to improve the verification efficiency. In addition, we propose a gas-efficient index maintenance for ADS updates on the blockchain. Finally, we provide a theoretical analysis of the complexity and verification guarantees of the proposed authenticated approaches and conduct extensive experiments on both real and synthetic datasets to demonstrate the efficient performance of our approaches.

In the domain of route planning, the critical concern for travel efficiency has shifted towards optimizing travel time over distance due to the rise in congestion and the dynamic nature of modern road networks. Addressing this shift, we introduce Dynamic Route Optimization with Multi-Category Constraints (DROMC) for Point-of-Interest (POI) visits, which seeks to find the most time-efficient path that satisfies a user-defined list of requirements, considering both the spatial and temporal dimensions. This paper proposes a novel approach that leverages a path enumeration algorithm, which iteratively validates the k-fastest paths until all user queries are satisfied, ensuring adherence to time constraints and POI availability. To enhance the algorithm's efficiency, we employ several key methodologies. First, we adapt the kSP algorithm to account for POI-focused path enumeration. We also introduce a Shared Prefix Tree (SPFT) with binary encoding, which significantly improves the storage and retrieval of path information. Moreover, we integrate a grid-based heuristic for quicker computation and implement strategic pruning methods to circumvent redundant calculations and manage POI business hours effectively. Our extensive experiments on real-world networks demonstrate the algorithm's superiority in finding more efficient paths in shorter time frames compared to existing methods.

Shortest distance computation is a fundamental operation in graph-related applications, especially in location-based services. The most efficient method is hop-labeling, which can answer queries in microseconds. However, when the traffic condition changes dynamically, they need a long time to maintain or an even longer time to re-construct, making it hard to catch up with numerous or frequent updates. As a result, real-life applications still rely on slow graph searching algorithms. To improve the hop labeling construction efficiency, we resort to GPU for its high parallelism power and propose the G2H index. Specifically, we first analyze the relation of the graph partitions, index performance, and parallelism to identify the most suitable partition scheme for G2H, with a hybrid scheme and optimized node ordering for faster contraction. Then, we propose a label-pruning method to reduce the label construction workload with several strategies designed to balance and improve the parallel label construction. Finally, experiments on real-life networks show that our G2H can finish construction within seconds for large urban networks and under one minute for large region networks with 6M vertices, which is several times faster than the state-of-the-art methods. Besides, G2H can answer hundreds of millions of queries per second, achieving two orders of magnitude acceleration.

In the field of educational data mining, Knowledge Tracing (KT) serves as a core technology for online learning systems, enabling personalized learning through dynamic modeling and real-time updates of learners' knowledge states. While existing research primarily focuses on assessing knowledge-point-level mastery, real-world educational contexts are characterized by substantial heterogeneity among students in terms of their cognitive ability and efficiency of knowledge acquisition. To address this, this article proposes an Ability-Enhanced Knowledge Tracing (AEKT) model under the Outcome-Based Education (OBE) theoretical framework, which integrates knowledge mastery with ability development. The AEKT model comprises four key components: a student ability acquisition module, an ability-enhanced knowledge assessment module, a foundational knowledge acquisition module, and a predictive module. Leveraging a Long Short-Term Memory (LSTM)-based deep learning architecture, AEKT extracts ability-specific features and knowledge hierarchy relevant to target problems, while dynamically modeling knowledge points, cognitive abilities, and the ability-aware knowledge acquisition process. Final student performance predictions are interpretable outputs constructed using an IRT-based model. The experimental results indicate that our proposed approach achieves superior performance compared to classical baseline methods on all four real-world public datasets. Ablation studies further validate the effectiveness of each module.

Cu2Se has become a highly promising thermoelectric material owing to its abundant elemental reserves, nontoxicity, and low thermal conductivity. In this work, flexible Cu2Se1-xIx (x = 0, 0.02, 0.04, and 0.06) films were synthesized by a cost-effective and facile hydrothermal approach, followed by vacuum-assisted filtration and hot pressing, and the films were dense with well-crystallized Cu2Se grains. The introduction of iodine provides an additional electron, acting as N-type doping and reducing carrier concentration. With the increase of x, the films exhibit a gradual reduction in electrical conductivity but a concurrent increase in the Seebeck coefficient. As a result, the Cu2Se1-xIx film with x = 0.04 shows an optimal power factor of ∼566.9 μW m-1 K-2 along with remarkable flexibility at room temperature. In addition, the maximum output power of the six-leg flexible thermoelectric generator fabricated using the Cu2Se0.96I0.04 film at a temperature difference of 22.8 K is 1.41 μW, with a power density of 4.02 W m-2. This work presents an effective strategy for developing high-performance flexible Cu2Se-based thermoelectric films.

Lithium metal batteries (LMBs) with solid polymer electrolytes (SPEs) offer higher energy density and enhance safety compared to the Li-ion batteries that use a graphite anode and organic electrolytes. However, achieving long cycle life for LMBs while enabling the use of high-voltage cathodes required the compatibility between cathode-SPE, rather than focusing solely on the individual components. This study presente a dual-functional poly(ionic liquid) (PolyIL)-based material that simultaneously serves as an SPE matrix and a cathode binder, constructing a cathode-SPE interface with exceptional (electro)chemical compatibility owing to the high ionic conductivity and wide electrochemical stability window. Additionally, a modified cellulose acetate (CA)-based PolyIL substrate, enriched with C═O and ─OH groups, is designed rationally and incorporated to assist the Li+ migration, leveraging their highly negative charge, and enhancing the mechanical strength of the SPE. Furthermore, an in situ polymerization approach is employed to assemble the cells, improving the physical compatibility at the cathode-SPE interface. As a result, the Li||LFP cell demonstrate stable cycling beyond 1100 cycles, and the Li||NCM811 cell reliably operates at a high cut-off voltage of up to 4.8 V.

Given a source, destination, and required keywords, the Optimal Route with Collective Spatial Keywords (ORCSK) query aims to find the shortest route covering all keywords. Existing Point of Interest (POI) candidate set-based and path expansion-based methods frequently produce inferior route quality or excessive time overhead, particularly under large-scale query keywords. To address this challenge, we introduce the DO framework, which pioneers the employ Deep Reinforcement Learning for the ORCSK. Specifically, DO first integrates the spatial index with the H2H index to generate and refine high-quality candidate sets. Subsequently, DO utilizes a Transformer-based model to determine the optimal route from the sets. To effectively combine spatial distance and POI attributes, we propose a novel dual-cross encoder architecture. Furthermore, leveraging this architecture, we introduce a multi-route generating strategy, exploiting parallel computing to enhance route quality. Our experiments on real-life road networks demonstrate superior route quality and response time compared to the state-of-the-art method, with an average improvement of 1-2 orders of magnitude in response time, and maintain high efficiency even under large-scale query keywords or dynamic POI attributes scenarios.

Efficient and stable lithium-ion batteries (LIBs) have garnered considerable attention; yet, the development of anode electrode materials continues to pose substantial challenges. While CoO electrode material boasts an ideal specific theoretical capacity, it is not without drawbacks, including significant volume expansion and concerns over safety performance, which hinder its viability as an anode material. In this research, we synthesized CoO/Co3O4 through a straightforward secondary hydrothermal treatment that locally oxidizes CoO, simultaneously creating oxygen vacancies. The incorporation of oxygen vacancies enhances the material’s internal conductivity and expedites the diffusion of electrons and ions, culminating in superior rate performance. Furthermore, the heterojunction structure diminishes the diffusion barrier, significantly enhancing the electrode’s reaction kinetics and overall electrochemical performance. At a modest current density of 0.1 A g−1, the CoO/Co3O4 composite demonstrates enhanced cycling stability, delivering a capacity of 1022 mAh g−1 after 100 cycles. Remarkably, even at an elevated current density of 1 A g−1, it sustains a capacity of 768.8 mAh g−1 over 400 cycles. The method of creating oxygen vacancies via autoxidation may pave the way for the advancement of multivalent oxide anode materials.

Recently, the keyword-aware routing problem has been increasingly studied, which is to return the optimal route from the starting point s to the destination t , satisfying all the user-specified keyword requirements. Most existing solutions focus only on the travel cost (distance or time) of the route and do not take into account the number of stops required to meet the keyword service. However, parking more often can degrade the user experience, as finding a parking space in a city is difficult and time-consuming. In this paper, we study the Keywords and Stops aware Optimal Routes (KSOR) problem, which finds a set of non-dominated routes with minimum distance and minimum number of stops. It is non-trivial to search such routes because the newly added Point of Interest (POI) may reduce the stops when expanding the road network, which poses challenges for pruning the partial routes. To avoid this dilemma, we first utilize a spatial index to find the potential POIs around the shortest path from s to t . To efficiently combine these POIs for generating valid routes, we propose two route generation methods. The light enumeration method prune the range of enumerated route combinations by the calculated lower and upper bounds of the stops. And the weighted calculation method generates routes by calculating the optimal combination of POIs under different weights of distance and stops. Extensive experiments conducted on real road networks show that the proposed methods are more efficient.

How to avoid congested roads and spend the least time to reach the destination has become an urgent requirement for citizens. When large-scale origin-destination queries arrive, these queries themselves will affect road congestion. Therefore, it is necessary to plan different routes in advance for these queries to avoid simultaneous occupation of the same roads, thereby reducing congestion and global travel time. Nevertheless, this is not trivial, as calculating which queries will occupy the same road resources is time-consuming. Moreover, re-planning the routes can also affect other queries, and this impact will spread. Existing methods either adopt greedy strategy to consider only the surrounding road condition, or the result quality is poor in the case of dense query. In this paper, we utilize the structural nature of the H2H index, to quickly find potentially conflicting queries and group them together. In this way, only the conflicts within each group need to be resolved in a local area, so the efficiency is improved. Additionally, several alternative paths are calculated and the global optimal routes are found in finite iterations. Extensive experiments conducted on real-world networks with synthetic and real workloads show that our method can provide higher quality results more efficiently.

Federated learning (FL) is a distributed machine learning method that enables multiple participants to contribute a well performed global model while their private training data remains in local devices. FL is promising in the edge computing system which has a large corpus of decentralized data and requires data privacy. However, traditional FL algorithms perform poorly with not independently and identically distribution data, especially highly skewed class imbalanced datasets. When solving class imbalance problems in FL, it is necessary to have prior knowledge of data distribution information, which cannot protect data distribution privacy. To fully protect privacy, we build a privacy protection bottom-up hierarchical federated learning (FedPBH) framework, which alleviates the imbalances by 1) Data sampling based on global data distribution, and 2) Bottom-up client participation. The proposed framework relieves global imbalance by data sampling based on the global data distribution which is obtained through privacy protection collaborative data distribution evaluation. For averaging the local imbalance, the proposed method creates bottom-up client participation, and these clients in the same local server asynchronously train their models. Experiments demonstrate that our FedPBH model provides full privacy protection with high classification performance.

Internet of Things (IoT) data provides rich data sources and application scenarios for trajectory representation learning. Trajectory representation learning aims to transform the original trajectory information into a general low-dimensional vector representation for many different downstream tasks (trajectory similarity calculation, anomaly detection, etc.). Current road network-based trajectory learning methods mainly focus on the spatial structure of the road network and often ignore the semantic information and complex feature information embedded in IOT data, in addition, the spatial and semantic properties of trajectories cannot be adequately preserved simultaneously. To this end, we propose a Trajectory Representation Learning framework based on Road network-TRLR. It first uses a graph attention network to learn the topological and semantic properties of road network segments, then it fuses node vectors and segment vectors as representations of node units, this is to support more data input types and enhanced node characteristics. Lastly, it learns the travel semantics of the trajectories through an information-enhanced transformer model, which captures the sequence information in the trajectory and generates the trajectory representation vector. In addition, we also propose four data augmentation methods to ensure that the trajectories can maintain both their spatial and semantic properties. To validate the effectiveness of our modeling approach, we conduct experiments on real datasets for similar trajectory search and mask prediction tasks. The experimental results demonstrate the performance improvement of our model.