Beyond Text and Tables: Vision-Language Model Integration in ComProScanner for Extracting Materials Data from Scientific Figures with High Accuracy

This feed aggregates recent CS/IR arXiv papers (June 2026) on multimodal information extraction, recommendation systems, and retrieval/RAG improvements. Key contributions include: (1) ComProScanner extended with a vision-language FigureExtractor/VLM pipeline to recover numeric composition–property pairs from scientific charts; (2) SentimentLens for aspect-based sentiment extraction and cross-modal reconciliation of text vs. ratings; (3) multimodal LLM-based music recommendation and a large benchmark; (4) SCALR, a synthetic cross-domain event generator to augment recommender training; (5)Prefix