Auditing Multimodal Claims: A Controlled Reproducibility Study of Graph-based Conversational Recommendation
Multimodal conversational recommender systems (CRS) integrate dialogue, knowledge graphs, and visual signals. Recent work reports substantial gains over the knowledge-graph baseline KGSF on the INSPIRED dataset (R@10: 0.132→0.257). However, this pipeline couples ViT-based image embeddings with zero-fill handling for 23...