Sequential recommendation models temporal patterns in user interaction sequences to capture dynamic preference changes. However, real-world user interaction data suffer from sparsity, hindering effective preference learning and limiting sequential recommendation model performance. Although existing studies employ time interval-aware data augmentation to address data sparsity, they inadequately mitigate popularity bias, leading to recommendations containing too many popular items. Consequently, this article proposes popularity bias mitigation based on time interval-aware data augmentation (TiPBMRec), which is described as a two-stage framework. In the first stage, TiPBMRec augments user interaction sequences using an item reshaper and a sequence refiner, dynamically generating augmented sequences. In the second stage, the augmented sequences are used to construct the time interval-aware dual-view graph and dual-channel conformity weight network, effectively capturing the changing patterns of user preferences. Experiments on real-world datasets demonstrate that TiPBMRec might be more optimal and achieve the popularity bias mitigation. This study provides a novel idea for exploring the combination of time interval-aware data augmentation and popularity bias mitigation. The related methods and conclusions might be valuable for enabling adaptive sequential recommendations.

To improve the particle swarm optimization algorithm's population diversity and global search ability, this study proposes a novel particle swarm optimization algorithm for meta-heuristic analysis mechanism based on population learning strategies and adaptive selection of leadership particles (ASLPSO) to achieve the fusion of population learning strategy and adaptive selection method of leadership particles. The particle swarm is adaptively separated into several populations through density peak clustering. Meanwhile, a new learning strategy is designed for analyzing the local optimal particles of each subgroup, that might enable the ordinary particles to learn effectively. The global optimum is relatively ensured by comparing the fitness value of each local optimal particle, which might obtain the best performances of multi-peak reference functions. After comparing with the approximate algorithms on multiple benchmark functions, it is found that the standard deviation and error mean value are improved. Therefore, ASLPSO has enhanced the population diversity and global search ability, prevented the algorithm from falling into premature too early.

Deep reinforcement learning, with its extensive applications and remarkable performance, is emerging as a pivotal technology garnering researchers’ attention. During the training process, there are frequent interaction and data exchange between agents and the environment, and the interaction information during training is closely tied to the training environment. Consequently, this process introduces a high risk of environmental privacy leakage. Malicious third parties may potentially steal state transition matrix or environmental information about the application domain of agent training, resulting in the compromise of user privacy. To address this issue, we propose novel differentially private value-based and policy-based deep reinforcement learning algorithms. Our methods have an advantage of being adaptable to various environmental privacy concerns. We also evaluate them in a customized experimental environment. Comparative experiments are conducted between the original and differentially private versions of the algorithms. The results indicate that our proposed approach can provide differential privacy protection to environmental information with minimal impact on algorithm performance, ultimately achieving a good balance between privacy and utility.