{"id":"W6931665716","doi":"10.5683/sp3/hskjey","title":"Dataset for Benchmarking Aspect Level Sentiment Classification on Reviews Requiring Coreference Resolution","year":2022,"lang":"en","type":"dataset","venue":"Borealis","topic":"Advancements in Transdermal Drug Delivery","field":"Pharmacology, Toxicology and Pharmaceutics","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Coreference; Benchmarking; Benchmark (surveying); Resolution (logic); Sentence","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":[0.001542422,0.002541117,0.000831344,0.00423373,0.001104626,0.001435678,0.0020458,0.00238951,0.01098299],"category_scores_gemma":[0.006008777,0.0003844077,0.001187222,0.00334194,0.000429486,0.001039946,0.00153941,0.001674513,0.02050142],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00203705,"about_ca_system_score_gemma":0.001848923,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02283551,"about_ca_topic_score_gemma":0.06332126,"domain_scores_codex":[0.9980818,0.0004672279,0.0002248332,0.000413245,0.0006413184,0.0001715993],"domain_scores_gemma":[0.9961922,0.001011504,0.0004513371,0.0006543125,0.001368258,0.0003222544],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0002015667,0.0002025599,0.002570614,0.0009698441,0.0001005523,0.0001264831,0.00007603434,0.001236968,0.0014628,0.0007374391,0.9755353,0.01677967],"study_design_scores_gemma":[0.0005214058,0.0001900236,0.0245501,0.0004418642,0.0001166285,0.0006602039,0.0003522747,0.01886626,0.005225156,0.001888326,0.9470447,0.0001429948],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"dataset","genre_gemma":"dataset","genre_scores_codex":[0.007071389,0.0009140302,0.001194732,0.0004462517,0.000245342,0.0002090753,0.9817245,0.003024997,0.005169573],"genre_scores_gemma":[0.003234484,0.0001088538,0.002586043,0.0001086853,0.00002474683,0.0001740826,0.9922411,0.00007755887,0.001444306],"genre_candidate":"dataset","genre_consensus":"dataset","teacher_disagreement_score":0.02283551,"threshold_uncertainty_score":0.04540515,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4339240540880192,"score_gpt":0.496446045824358,"score_spread":0.06252199173633882,"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."}}