Multimedia content offers additional context for recommender systems to better understand user interests. Existing studies on multimodal recommendation primarily focus on constructing item-item semantic graphs. However, most of these methods capture only shallow semantic structures based on feature similarity and struggle to model more complex or cross-entity semantic relationships (e.g., user-item). Moreover, in these methods, collaborative signals often dominate and suppress semantic knowledge, which limits its role in representation learning. To address these issues, we propose SCALE, a novel framework that combines subspace-aware graph construction and contrastive alignment for multimodal recommendation with large language models. Specifically, we first use large language models and encoders to extract user and item features. Following the subspace clustering assumption, we apply the Orthogonal Matching Pursuit algorithm to mine complex semantic structures within the item-item, user-user, and user-item spaces, and integrate them into a unified semantic graph. We then perform graph convolution on both the semantic and interaction graphs, and aggregate the results for recommendation. Furthermore, contrastive losses are employed to enhance semantic fusion and alignment. Extensive experiments on five real-world datasets demonstrate that SCALE significantly outperforms state-of-the-art multimodal recommendation models, highlighting its effectiveness in modeling complex relationships and integrating semantic knowledge with collaborative signals.

Jinze Li, Yang Zhang, Xin Yang, Jiayi QU, Jinfeng Xu, Shuo Yang, Junhua Ding, Edith Cheuk-Han Ngai. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Recently, patch-based transformer methods have demonstrated strong effectiveness in time series forecasting. However, the complexity of self-attention imposes demands on memory and compute resources. In addition, though patches can capture comprehensive temporal information while preserving locality, temporal information within patches remains important for time series prediction. The existing methods mainly focus on modeling long-term dependencies across patches, while paying little attention to the short-term dependencies within patches. In this paper, we propose the Global and Local Frequency-domain Network (GLFNet), a novel architecture that efficiently learns global time dependencies and local time relationships in the frequency domain. Specifically, we design a frequency filtering layer to learn the temporal interactions instead of self-attention. Then we devise a dual filtering block consisting of global filter block and local filter block which learns the global dependencies across patches and local dependencies within patches. Experiments on seven benchmark datasets demonstrate that our approach achieve superior performance with improved efficiency.