"HomBreak"

"This dataset collection provides three multimodal session-based recommendation benchmarks, including Cellphones, Sports, and Grocery. The datasets are constructed from real-world user-item interaction records and contain session-level interaction sequences together with rich multimodal item information, including textual descriptions and visual features. Each dataset is carefully organized to preserve the sequential characteristics of user behaviors while providing aligned multimodal signals for item representation learning and preference modeling.These datasets are designed to support research on multimodal session-based recommendation, multimodal representation learning, and diversity-aware recommendation. They provide a comprehensive evaluation platform for studying how different modalities contribute to user preference understanding, next-item prediction, and recommendation diversity enhancement. By covering multiple domains with distinct item characteristics and user behavior patterns, this dataset collection enables researchers to investigate the effectiveness and generalizability of multimodal recommendation approaches across different scenarios.Furthermore, the aligned multimodal information and session interaction data facilitate the development and evaluation of advanced recommendation techniques, including multimodal fusion, modality interaction modeling, preference disentanglement, and diversity-oriented recommendation. This collection aims to provide a publicly available benchmark resource for the research community and promote further exploration of how multimodal signals can be effectively leveraged to improve personalized recommendation systems."