Controlled Comparison of KAN and MLP Fusion Heads in Multimodal Sentiment Analysis — data, results and code

Fixed MVSA-Multiple test split (1,701 pairs), step-by-step filtering record (19,600 → 17,027), per-seed metrics of all training runs, single-session latency measurements, statistical summaries (seed-paired differences, 90% CI, TOST) and the training/analysis code accompanying the article "Controlled Comparison of KAN and MLP Fusion Heads in Multimodal Sentiment Analysis" (Electronics, MDPI). The datasets themselves (MASAD, MVSA-Multiple) are not redistributed; see README.md. Data and result tables: CC BY 4.0; code (kan_mlp_fusion_code.zip): MIT.