In e-commerce and online content platforms, user behaviors typically exhibit multi-scale characteristics, including long-term preferences, periodic habits, and short-term responses, while users’ decision-making processes heavily rely on multimodal item content. However, existing methods often focus on a single temporal scale or consider multimodal information only as independent features, making it difficult to achieve effective collaborative modeling under multi-scale dynamics. To address this issue, we propose a Cross-View Collaborative Recommendation with Multimodal and Multi-Scale User Behaviors (MM-MCSRec). Specifically, we first construct multi-view heterogeneous graphs based on brand, category, time, and price. Then, spectral filters are employed to decompose user behavioral signals into three frequency bands: long-term preferences, periodic habits, and short-term responses, while a node-level gating mechanism is introduced to enhance feature selectivity. Furthermore, periodic modulation and dynamic reweighting strategies are designed in the time and price views to better capture periodic patterns and short-term responses. Experiments on multiple real-world datasets demonstrate that MM-MCSRec outperforms existing recommendation methods.

With the rapid development of e-Commerce, product recommendation systems play a crucial role in enhancing user experience and increasing the volume of transaction on the platform. However, existing recommendation systems generally fail to fully consider the fine-grained features of products and primarily focus on users' positive preference features while neglecting potential negative preference features. This limitation constrains the accuracy and diversity of recommendation systems. To address this, we propose a novel multimedia recommendation model called ''Temporal Causal Fine-grained Recommendation'' (TCFRec). Specifically, we first perform fine-grained feature extraction and classification of products based on the CLIP model and a multi-level complementary attention mechanism. Subsequently, we leverage a personalized time decay strategy and causal contrastive learning to deeply explore both users' positive and negative preferences. Furthermore, counterfactual reasoning is utilized to identify and eliminate spurious correlations in multimodal features. Finally, by integrating users' fine-grained positive preference classification, negative preference classification, and the influence of social networks, we achieve accurate and diversified personalized recommendations. We conducted extensive experiments to verify the effectiveness and rationality of TCFRec.

Characterizing human driver's driving behaviors from GPS trajectories is an important yet challenging trajectory mining task. Previous works heavily rely on high-quality GPS data to learn such driving style representations through deep neural networks. However, they have overlooked the driving contexts that greatly govern drivers' driving activities and the data sparsity issue of practical GPS trajectories collected at a low-sampling rate. Besides, existing works omit the cold start problem, where the newly joined drivers usually have insufficient data to learn accurate driving style representations. To address these limitations, we present an adversarial driving style representation learning approach, named$\mathtt {Radar}$. In addition to summarizing statistic features from raw GPS data,$\mathtt {Radar}$also extracts contextual features from three aspects of road condition, geographic semantic, and traffic condition. We exploit the advanced semi-supervised generative adversarial networks to construct our learning model. By jointly considering statistic features and contextual features, the trained model is able to efficiently learn driving style representations from practical GPS trajectory data. Furthermore, we enhance$\mathtt {Radar}$'s representation learning for drivers owning limited training data with some basic data augmentation strategies and a novel auxiliary driver based data augmentation method. Experiments on two benchmark applications,i.e., driver identification and driver number estimation, with a large real-world GPS trajectory dataset demonstrate that$\mathtt {Radar}$can outperform the state-of-the-art approaches by learning more effective and accurate driving style representations.