{"id":"W2113376122","doi":"","title":"Analysis of Summarization Evaluation Experiments","year":2007,"lang":"en","type":"article","venue":"North American Chapter of the Association for Computational Linguistics","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université Laval","funders":"","keywords":"Automatic summarization; Computer science; Terminology; Presentation (obstetrics); Information retrieval; Focus (optics); Natural language processing; Multi-document summarization; Artificial intelligence; Data mining; Linguistics","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0008919789,0.00008414971,0.0002116824,0.0002721515,0.00008093388,0.0000203432,0.0004237854,0.00002928284,0.000001376085],"category_scores_gemma":[0.003227034,0.00007381995,0.0001503686,0.001241028,0.00004864323,0.00004651422,0.00008159491,0.00005460589,2.562658e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001750834,"about_ca_system_score_gemma":0.00006963658,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004843674,"about_ca_topic_score_gemma":0.00006407699,"domain_scores_codex":[0.9984518,0.00003966277,0.0004351578,0.0001702249,0.0007798002,0.0001234013],"domain_scores_gemma":[0.995104,0.0006372508,0.001362905,0.0002007279,0.002668974,0.00002619079],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00003771843,0.0001436644,0.6425328,0.00002522323,0.0009290022,1.882425e-7,0.001058665,0.08612062,0.0001590794,0.2339198,0.00008498508,0.03498821],"study_design_scores_gemma":[0.000342291,0.00009870698,0.3441454,0.00001492745,0.0005915285,1.488514e-7,0.00002599698,0.6301433,0.004707769,0.01922524,0.0005109159,0.0001937479],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02920704,0.00005572589,0.9695857,0.00008836086,0.0002905144,0.0002843828,0.00005258089,0.00005123351,0.000384495],"genre_scores_gemma":[0.8030269,7.942743e-7,0.1967296,0.00009717061,0.00004525765,0.000004892941,0.0000730097,0.000004881045,0.00001742402],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.7738199,"threshold_uncertainty_score":0.3863294,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01988773488424193,"score_gpt":0.3238424057267155,"score_spread":0.3039546708424736,"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."}}