{"id":"W2200523128","doi":"","title":"Evaluating Information Retrieval Systems With Multiple Non-Expert Assessors","year":2013,"lang":"en","type":"dissertation","venue":"UWSpace (University of Waterloo)","topic":"Mobile Crowdsensing and Crowdsourcing","field":"Computer Science","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"University of Waterloo","keywords":"Ranking (information retrieval); Relevance (law); Computer science; Information retrieval; Key (lock); Expert system; Test (biology); Machine learning; Data mining; Artificial intelligence","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.000374072,0.0002920786,0.0004405237,0.0003576055,0.0003744744,0.0002926272,0.000777552,0.0002993208,0.00001790339],"category_scores_gemma":[0.00004091423,0.0002943102,0.0001097878,0.0003977961,0.00005149474,0.00181746,0.0001049673,0.000305968,0.0001040072],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001295433,"about_ca_system_score_gemma":0.0002139675,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03788892,"about_ca_topic_score_gemma":0.002053281,"domain_scores_codex":[0.9982414,0.00009059373,0.0002233264,0.0003842447,0.0007168442,0.0003436105],"domain_scores_gemma":[0.9978192,0.00008188348,0.0006327023,0.0006520492,0.0006921798,0.0001219512],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0007153233,0.0001947377,0.001110387,0.002478955,0.0005745806,0.00008456019,0.8899861,0.02179502,0.03463676,0.0006802349,0.007030057,0.04071329],"study_design_scores_gemma":[0.002227738,0.0007169239,0.005228753,0.001896533,0.0001295817,0.00002613729,0.2762477,0.705093,0.006468981,0.00001815195,0.0008326329,0.001113979],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9914827,0.00006130369,0.006443655,0.0001130564,0.0006477332,0.0005293145,0.000005673076,0.000159184,0.0005573991],"genre_scores_gemma":[0.8330907,0.00004507701,0.03270934,0.00002695722,0.00008749635,0.00000201481,0.0004329914,0.00003824553,0.1335672],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6832979,"threshold_uncertainty_score":0.9999509,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01371373316032038,"score_gpt":0.2263860097607907,"score_spread":0.2126722766004703,"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."}}