{"id":"W4408173301","doi":"10.26434/chemrxiv-2025-8z6h2","title":"MERMaid: Universal multimodal mining of chemical reactions from PDFs using vision-language models","year":2025,"lang":"en","type":"preprint","venue":"ChemRxiv","topic":"Advanced Text Analysis Techniques","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Vector Institute; Canadian Institute for Advanced Research; University of Toronto","funders":"Canada First Research Excellence Fund; Canadian Institute for Advanced Research","keywords":"Computer science; Artificial intelligence","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0001604733,0.000340494,0.0006147451,0.0003472473,0.00006317088,0.00006814103,0.001684913,0.0003998325,0.00002156886],"category_scores_gemma":[0.0000770283,0.0003785867,0.0003283803,0.0004709935,0.0001221834,0.0004901461,0.002658531,0.0005964676,0.000002140359],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002570295,"about_ca_system_score_gemma":0.0002469129,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007214839,"about_ca_topic_score_gemma":0.000004995698,"domain_scores_codex":[0.9978349,0.00005558502,0.0005466661,0.0009261899,0.000361117,0.0002755444],"domain_scores_gemma":[0.997427,0.0002311164,0.0004726953,0.001577538,0.0001913677,0.0001002978],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00002852959,0.000366,0.000104488,0.0002204809,0.0005519179,0.00006798386,0.005043505,0.05384408,0.9117289,0.005602242,0.0003906399,0.02205127],"study_design_scores_gemma":[0.0001410897,0.000004515026,0.00000784171,0.0003274141,0.00007790221,0.000001245474,0.00009301039,0.6951452,0.2958415,0.008056062,0.00004101186,0.0002631648],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2313834,0.0002222253,0.7663221,0.0000594164,0.0001509711,0.0001425108,0.00001626688,0.0003408315,0.001362235],"genre_scores_gemma":[0.5382169,0.00002042946,0.4615424,0.00002254589,0.00005304786,0.000007670607,0.00003825809,0.00001216874,0.00008650371],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6413012,"threshold_uncertainty_score":0.9998666,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02439863093086714,"score_gpt":0.3166167865368164,"score_spread":0.2922181556059493,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}