{"id":"W3200891190","doi":"","title":"Inspecting the Factuality of Hallucinated Entities in Abstractive Summarization.","year":2021,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Topic Modeling","field":"Computer Science","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Hallucinating; Automatic summarization; Computer science; Natural language processing; Artificial intelligence; Information retrieval","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":[0.002103678,0.001065855,0.0004906635,0.002334013,0.0004557964,0.001601113,0.0008410392,0.0008878334,0.001532503],"category_scores_gemma":[0.01400213,0.0003040887,0.0005174853,0.001058727,0.0004255778,0.002604953,0.00115266,0.0009963129,0.001181445],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004090902,"about_ca_system_score_gemma":0.0004270406,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001518304,"about_ca_topic_score_gemma":0.003108074,"domain_scores_codex":[0.9990583,0.0003219053,0.00008754721,0.0002329153,0.0002423479,0.00005701056],"domain_scores_gemma":[0.9912933,0.005198578,0.001084648,0.0008715666,0.001356119,0.0001958017],"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":[0.0008475952,0.0001524561,0.01512337,0.001489125,0.0003633965,0.0008645636,0.002591572,0.03319453,0.1108193,0.007821875,0.02058956,0.8061426],"study_design_scores_gemma":[0.00008119185,0.0007137981,0.02825134,0.0002313788,0.0005271385,0.001325531,0.001580555,0.7693487,0.1287093,0.03155526,0.03751811,0.0001577012],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1210899,0.003133755,0.862231,0.0008710632,0.0002655369,0.0002510289,0.002370113,0.006231803,0.003555776],"genre_scores_gemma":[0.71468,0.001174405,0.2726793,0.0001882896,0.0003419922,0.0001300676,0.006439168,0.000389009,0.003977868],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.002334013,"threshold_uncertainty_score":0.01112539,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09452848709654141,"score_gpt":0.1948844294690246,"score_spread":0.1003559423724832,"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."}}