{"id":"W1999302328","doi":"10.1186/2041-1480-5-11","title":"Benchmarking infrastructure for mutation text mining","year":2014,"lang":"en","type":"article","venue":"Journal of Biomedical Semantics","topic":"Biomedical Text Mining and Ontologies","field":"Biochemistry, Genetics and Molecular Biology","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of New Brunswick","funders":"New Brunswick Innovation Foundation","keywords":"Computer science; Benchmarking; SPARQL; Information retrieval; Ontology; RDF; Data mining; Annotation; Benchmark (surveying); Data science; Semantic Web; 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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02477304,0.001731658,0.001469938,0.009684004,0.002316749,0.00388058,0.005354342,0.001730177,0.007214346],"category_scores_gemma":[0.0519265,0.0007507184,0.001583739,0.01017399,0.001351353,0.009203847,0.005846601,0.002095902,0.005492264],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003956407,"about_ca_system_score_gemma":0.006854047,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006786299,"about_ca_topic_score_gemma":0.003572321,"domain_scores_codex":[0.9754655,0.007200564,0.004746299,0.003754661,0.007349181,0.001483853],"domain_scores_gemma":[0.9505344,0.011941,0.003514657,0.01644846,0.01547604,0.002085436],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002763261,0.001988039,0.02523665,0.004649151,0.0005473192,0.001294668,0.002468098,0.05939538,0.04418221,0.08209286,0.233321,0.5420614],"study_design_scores_gemma":[0.0005180062,0.00106016,0.02299277,0.001148541,0.0003536688,0.001355556,0.00144972,0.3993921,0.1278121,0.08909531,0.354299,0.0005230621],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.05174452,0.001282297,0.6453143,0.002481823,0.0003989683,0.002905316,0.04321505,0.2274861,0.02517151],"genre_scores_gemma":[0.2474071,0.0009342615,0.4541871,0.0007053171,0.0001668977,0.003438206,0.2736392,0.0149525,0.00456938],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.02477304,"threshold_uncertainty_score":0.1310139,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.008037719161641739,"score_gpt":0.2664438411905976,"score_spread":0.2584061220289558,"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."}}