In resource-constrained environments such as IoT sensors and mobile devices, there is a strong demand for efficient conjunctive keyword search over privacy-sensitive data. However, existing schemes struggle to simultaneously suppress sterm equality leakage, the cross-query intersection pattern (IP), and the volume pattern without incurring prohibitive overhead. In this paper, we present XORCMM, a practical volume-hiding encrypted conjunctive multi-map (EMM) designed for robust leakage suppression. First, we shift the index construction from single keywords to global-ordering co-occurrence pairs, which ensures that search tokens are no longer tied to static keyword identities, thereby suppressing sterm equality leakage. Second, we integrate an incremental multiset hash aggregation mechanism directly into a fully padded Xor filter. This allows the server to aggregate multiple conjunctive results into a single, fixed length response, concealing both IP and volume patterns while eliminating the data redundancy of prior schemes. Third, we employ a prefix-constrained PRF to compactly encode keyword pairs, generating succinct query tokens whose size is independent of keyword volumes. Formal security analysis proves that XORCMM is adaptively secure with sterm equality, IP, and volume leakages hidden. Experimental results demonstrate that XORCMM achieves up to a 2.99× speedup in client setup, a 3.3× speedup in server query time, and reductions of 47% in response size and 84.61% in search token size, providing a stronger security guarantee with significantly higher efficiency.

Highly precise and rapid global-optimal merchandise price prediction can offer reference value for enterprises' strategic decisions, such as procurement and risk assessment in the economic market. This paper proposes a merchandise price prediction method based on improving Deep Neural Networks(DNN) using the Grey Wolf-Particle Swarm Algorithm. To address the gradient vanishing or explosion caused by the input of multi-feature initial parameters into deep neural networks in merchandise price prediction, this paper optimizes the weight and bias parameters of DNN by improving the Chaotic Theory of the Grey Wolf-Particle Swarm Optimization Algorithm (IC-GWO-PSO). The concept of circular mapping in the GWO algorithm is introduced to avoid the problem of the particle swarm being prone to local optima and achieve a jump search to obtain the global optimal solution of the parameters of DNN. Moreover, the tent mapping in chaos theory is adopted to control the adjustment speed of particles and prevent them from exceeding the solution boundary, enhancing the exploration efficiency and convergence accuracy of the algorithm in the solution space. Experimental results indicate that IC-GWO-PSO outperforms baseline methods such as GWO-PSO and IGWO-IPSO in procurement price prediction. The three indicators, RMSE, R, and Fitness, reach 2.49, 0.9937, and 1583, respectively, demonstrating its performance advantages in global rapid solution seeking and precise price prediction.

As a real-time data sensor in vehicle networks, dash cams capture video with both static (object positions, scene layouts) and dynamic (motion paths, speed changes) spatio-temporal data. Analyzing this information helps predict scene changes and prevent accidents. To improve spatio-temporal feature representation and prediction accuracy, we propose a traffic accident prediction model based on spatio-temporal depth maps and Transformer, named the STDMP model. The model creates 3D depth maps using monocular estimation, aligns spatial-temporal features through two-stream architecture, and integrates GRU's sequential processing with Transformer's attention mechanism for precise short-term tracking and long-term dependency modeling. Experimental data shows the STDMP significantly outperforms baseline models on the DAD dataset. The key metrics (AP:64.3, mTTA:4.25) confirm the effective combination of two techniques: depth-based spatial modeling and Transformer's temporal feature capture.

Urban public transportation, epitomized by buses, has become integral in facilitating efficient and convenient mobility for citizens. Despite its growth, the prevalent bus operation paradigm (static departure schedules, linear route planning, underutilized spatio-temporal data, etc.) often fails to cater to the fluid travel needs of passengers. This study presents a novel dynamic scheduling method based on spatial graph convolution and proximal policy optimization, called DSPTR. Combining a bus dynamic scheduling model (BDS) and a passenger path planning model (PRP), DSPTR method judiciously determines bus departure intervals and formulates passenger travel routes. Leveraging the Weight Graph Sample and AggreGatE (WGraphSAGE), DSPTR adeptly extract road network characteristic, ensuring the feature representation quality of bus resource states. The Proximal Policy Optimization (PPO) further refines the scheduling by optimizing bus multi-route departures and passenger route recommendations. Through experiments and evaluations on real data in Shanghai, DSPTR outperforms baselines such as DDPG and A2C. Through the traffic resource scheduling plan generated by DSPTR, bus operating costs and passenger travel times can be significantly reduced, highlighting its potential to improve the operational efficiency of the urban public transportation system.

Using tracking data obtained from the smartphone and Internet survey, a data-driven machine learning method is proposed to identify trip ends. In previous literature, this is usually done based on some predefined rules, which have been confirmed to be valid. Nonetheless, these rule-based methods largely depend on researchers' own knowledge, which is inevitably subjective and arbitrary. Moreover, they are not effective enough to process the huge amount of data in the era of big data. In this paper, millions of smartphone-based GPS tracking data are targeted. A group of attributes, such as travel speed, distance, and heading, are derived to characterize the smartphone holders' travel status. In other words, the tracking points could be identified as being at the state of traveling or non-traveling, based on which the trip ends are easily detected. In contrast to those rule-based methods, a random forest is utilized in this paper as the classification model, with no subjective rules predefined for classification. This data-driven model is automatically built. The results show that after training the GPS tracking data of 1393 days and the prompted recall (PR) survey data using the random forest, the accuracy of trip ends identification on tracking data of 697 days is 96.17%. The current analysis is free from personal experiences, which is expected to be useful for the smartphone-based survey data in the era of big data.