{"id":"W1988220974","doi":"10.3115/1654679.1654682","title":"Challenges in evaluating summaries of short stories","year":2006,"lang":"en","type":"article","venue":"","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"","keywords":"Computer science; Sentence; Quality (philosophy); Natural language processing; Information retrieval; Artificial intelligence; Data science; Epistemology","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":[0.02631622,0.001152079,0.001756924,0.002947395,0.001097719,0.004481045,0.001860067,0.002172076,0.002093511],"category_scores_gemma":[0.1351774,0.0004708345,0.0007124932,0.002460411,0.0007730474,0.006402799,0.001754929,0.001169414,0.001401933],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000879011,"about_ca_system_score_gemma":0.0007728913,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001321743,"about_ca_topic_score_gemma":0.001769825,"domain_scores_codex":[0.9589949,0.02401963,0.006057788,0.003067729,0.007515104,0.0003449006],"domain_scores_gemma":[0.8134314,0.1480114,0.008529831,0.007470276,0.02059909,0.001957909],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.002162933,0.000681394,0.009402839,0.006886185,0.0007773179,0.0006921521,0.007969795,0.0171101,0.03838082,0.003648615,0.01572294,0.8965649],"study_design_scores_gemma":[0.002236403,0.01880039,0.0994631,0.003456842,0.002327197,0.006950312,0.03017653,0.3172752,0.2022733,0.05450153,0.2611459,0.001393329],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5753621,0.03510207,0.3518688,0.005406532,0.001134285,0.002243739,0.005158075,0.01022812,0.01349614],"genre_scores_gemma":[0.6643196,0.004749167,0.3179977,0.0005119473,0.0007500906,0.0008570103,0.007822552,0.0008996951,0.002092182],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02631622,"threshold_uncertainty_score":0.1391752,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07351163251175052,"score_gpt":0.3469771445445267,"score_spread":0.2734655120327761,"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."}}