Chanmi Kang, Bongguk Son, Junsu Cho · ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 2026
DOI: 10.1145/3799682.3841022
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
Multimodal recommender systems have attracted increasing attention for their potential to enhance recommendation performance by leveraging textual and visual item information. However, existing approaches often treat multimodal signals as auxiliary features for item-level discrimination, overlooking two intrinsic characteristics of real-world multimodal data: 1) noise and uneven informativeness across modalities and items, and 2) fine-grained intent-related semantics embedded in multimodal representations. In this paper, we propose GEMIA, a Generative Multimodal Intent modeling framework with Adaptive integration for recommendation, which infers target-side multimodal semantic representations related to user preferences which explicitly models intent-related multimodal semantics while adaptively integrating heterogeneous modality signals. GEMIA consists of two key components: a Multimodal Semantic Inference (MSI) module that adopts a generative representation-inference approach to generate the next item's text and image representations in a continuous multimodal semantic space, enabling the model to capture fine-grained intent-related information; and an Adaptive Multimodal Fusion (AMF) module that integrates multimodal representations via item-specific gating, accounting for item-dependent modality noise and uneven informativeness. Furthermore, GEMIA aggregates item-, text-, and image-based scores using item-wise weighting to flexibly adjust reliance on multimodal and item-level information. Extensive experiments on four real-world datasets demonstrate that GEMIA consistently outperforms state-of-the-art baselines, with more pronounced improvements in long-tail scenarios. Our code is available at https://github.com/kCMI113/GEMIA.
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 1 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
No comments yet — start the discussion below.