{"id":"W2287148974","doi":"10.1145/2888422.2888439","title":"Report on the SIGIR 2015 Workshop on Reproducibility, Inexplicability, and Generalizability of Results (RIGOR)","year":2016,"lang":"en","type":"article","venue":"ACM SIGIR Forum","topic":"Web Data Mining and Analysis","field":"Computer Science","cited_by":59,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"Università degli Studi di Padova","keywords":"Generalizability theory; Computer science; Presentation (obstetrics); Thursday; Information retrieval; Library science; Data science; Statistics; Medicine","routes":{"ca_aff":true,"ca_fund":false,"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":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4204637,0.004047017,0.00427287,0.007307969,0.007461053,0.02533596,0.008611387,0.009897928,0.03964162],"category_scores_gemma":[0.5271295,0.002115322,0.005136801,0.003938264,0.007451684,0.02916217,0.02470973,0.01356752,0.02502416],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007094688,"about_ca_system_score_gemma":0.01513458,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009034202,"about_ca_topic_score_gemma":0.006196078,"domain_scores_codex":[0.6590006,0.1876811,0.02007467,0.02524964,0.09905916,0.008935003],"domain_scores_gemma":[0.3206436,0.3235835,0.01790087,0.107462,0.2069122,0.02349778],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001918687,0.0007566003,0.00595755,0.002047982,0.0006152798,0.000409784,0.004064595,0.001457883,0.004444692,0.01352579,0.748414,0.2163873],"study_design_scores_gemma":[0.001091375,0.001509704,0.01334012,0.004033119,0.0008992302,0.0007777915,0.005072844,0.01136812,0.01424693,0.08823447,0.8586029,0.0008233882],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"methods","genre_gemma":"review","genre_scores_codex":[0.02831216,0.06595184,0.3325501,0.3050986,0.1459833,0.008227564,0.01420851,0.01117791,0.08849011],"genre_scores_gemma":[0.2615598,0.02334031,0.3211905,0.08744568,0.06198272,0.01338972,0.06458687,0.01743254,0.1490718],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.5795363,"threshold_uncertainty_score":0.7146714,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04259519423738254,"score_gpt":0.3006233049999692,"score_spread":0.2580281107625866,"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."}}