{"id":"W335269702","doi":"","title":"Measuring Robustness with First Relevant Score in the TREC 2012 Microblog Track","year":2012,"lang":"en","type":"article","venue":"Text REtrieval Conference","topic":"Information Retrieval and Search Behavior","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Open Text (Canada)","funders":"","keywords":"Microblogging; Robustness (evolution); Computer science; Social media; Rank (graph theory); Measure (data warehouse); Context (archaeology); Post hoc; Artificial intelligence; Information retrieval; Data mining; Mathematics; World Wide Web; 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":[],"consensus_categories":[],"category_scores_codex":[0.01907652,0.001619228,0.001669307,0.005394723,0.0009047722,0.002386232,0.001485196,0.002475959,0.001043061],"category_scores_gemma":[0.111375,0.0004722254,0.001291653,0.003882176,0.00108928,0.002947136,0.00121206,0.001844048,0.001149766],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001308768,"about_ca_system_score_gemma":0.001203851,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00478413,"about_ca_topic_score_gemma":0.004627989,"domain_scores_codex":[0.9692547,0.01126628,0.003714454,0.003509126,0.01133103,0.0009243749],"domain_scores_gemma":[0.8357006,0.1086738,0.01805716,0.0189492,0.01610937,0.002509939],"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":[0.02266791,0.006035715,0.2286641,0.005635448,0.007240327,0.0006787422,0.001293359,0.1009075,0.15254,0.002036334,0.02530118,0.4469995],"study_design_scores_gemma":[0.0008278381,0.0197634,0.4647777,0.0002198499,0.001731439,0.00204945,0.000650366,0.2417831,0.2562043,0.002427619,0.008666447,0.0008983365],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9556683,0.004107628,0.02256369,0.0004103993,0.0003284338,0.0006198015,0.005686607,0.00480454,0.005810445],"genre_scores_gemma":[0.9719864,0.0003007946,0.02119988,0.0001159993,0.0001623967,0.0002173168,0.004707912,0.0002422652,0.001067089],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01907652,"threshold_uncertainty_score":0.1008875,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08410569437123029,"score_gpt":0.2573694548114804,"score_spread":0.1732637604402501,"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."}}