{"id":"W2130998770","doi":"10.1145/1596517.1596518","title":"Extrinsic summarization evaluation","year":2009,"lang":"en","type":"article","venue":"ACM Transactions on Speech and Language Processing","topic":"Topic Modeling","field":"Computer Science","cited_by":30,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Sixth Framework Programme","keywords":"Automatic summarization; Computer science; Task (project management); Information retrieval; Audit; Human–computer interaction; Natural language processing; Artificial intelligence","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":[0.000240762,0.00008657066,0.0000743233,0.0001199614,0.0001905012,0.0001753455,0.0002264463,0.00004674392,0.00002094664],"category_scores_gemma":[0.0000343554,0.00008184119,0.00002080504,0.0002706592,0.000009944462,0.0005554298,0.00000434935,0.0001115634,0.00000529818],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003050285,"about_ca_system_score_gemma":0.00005190429,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001322994,"about_ca_topic_score_gemma":0.000009614182,"domain_scores_codex":[0.9991844,0.00003169585,0.0001266187,0.0002657591,0.0002523989,0.0001391285],"domain_scores_gemma":[0.9994972,0.00002264689,0.00004018678,0.0003234005,0.00006668054,0.00004992089],"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":[0.000002768269,0.00002653768,0.000008731467,0.00000650536,0.000001674052,0.000003006769,0.001136823,0.0003768467,0.001691835,0.0001115692,0.000001923205,0.9966318],"study_design_scores_gemma":[0.001052171,0.0001651109,0.001575978,0.0001418425,0.00004824456,0.00007983652,0.0007226022,0.9535074,0.03165402,0.01043249,0.0002147411,0.0004055425],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0472578,0.0008914804,0.9497743,0.001021055,0.00006092007,0.0001224506,4.852222e-7,0.000150835,0.0007206403],"genre_scores_gemma":[0.8828462,0.00001771019,0.1165879,0.000399891,0.00003319665,0.000005434797,0.000002088881,0.000003789747,0.0001037413],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9962263,"threshold_uncertainty_score":0.3337387,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02421893031646663,"score_gpt":0.2929102533355549,"score_spread":0.2686913230190883,"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."}}