{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06178143,0.001932587,0.001989486,0.004615422,0.001431557,0.002841177,0.001310339,0.001309015,0.00323246],"category_scores_gemma":[0.3287203,0.0005187626,0.001340213,0.004538119,0.00113637,0.003153387,0.001631688,0.001836597,0.001081524],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002336572,"about_ca_system_score_gemma":0.001320842,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008541035,"about_ca_topic_score_gemma":0.0007710963,"domain_scores_codex":[0.8534503,0.09847936,0.0158567,0.005768466,0.02467674,0.001768533],"domain_scores_gemma":[0.4434772,0.4412593,0.02582361,0.02193425,0.06543244,0.002073239],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.03346079,0.00812998,0.07484902,0.01575957,0.005518123,0.0007873436,0.01067213,0.03390468,0.08933765,0.009160733,0.04247938,0.6759406],"study_design_scores_gemma":[0.004653532,0.05362681,0.4646313,0.003656033,0.007907915,0.001699622,0.01037872,0.1222071,0.2367916,0.01900806,0.07400209,0.001437239],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9035753,0.005548877,0.05841694,0.001237951,0.0007223284,0.005571595,0.008591801,0.002683952,0.01365132],"genre_scores_gemma":[0.9487934,0.0007594679,0.03164917,0.0004049794,0.0002234583,0.005344893,0.009149862,0.000791805,0.002882986],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06178143,"threshold_uncertainty_score":0.3267354,"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."}}