{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001796361,0.0005364501,0.0006023944,0.000250529,0.0007631882,0.00003760742,0.0008920277,0.0004761619,0.01049427],"category_scores_gemma":[0.0002189213,0.0005618577,0.0001912652,0.0001851279,0.0001600232,0.0001553554,0.0002421273,0.001514326,0.00005962978],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000679646,"about_ca_system_score_gemma":0.0001335179,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007760433,"about_ca_topic_score_gemma":0.001216627,"domain_scores_codex":[0.9961779,0.000852754,0.0009198114,0.0009682329,0.0004426008,0.000638717],"domain_scores_gemma":[0.9973499,0.0005865822,0.0007687842,0.001041571,0.00007680456,0.0001763747],"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":[0.0005531129,0.0003465186,0.000005397814,0.0003932739,0.0001704775,0.00002385178,0.00003414735,0.0005651357,0.0001532338,0.0001941495,0.9810719,0.01648877],"study_design_scores_gemma":[0.0009686641,0.0002780569,0.00008760666,0.0001297938,0.000563475,0.00000854321,0.0000218175,0.001005712,0.0001472909,0.00009774371,0.9961601,0.000531121],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"dataset","genre_gemma":"dataset","genre_scores_codex":[0.000003958058,0.001249179,0.001423268,0.0002778945,0.001762634,0.002237837,0.9927316,0.00003782998,0.0002757469],"genre_scores_gemma":[0.00001445094,0.01247684,0.00055185,0.002737753,0.0006325517,0.001399437,0.9819829,0.00004265041,0.0001614931],"genre_candidate":"dataset","genre_consensus":"dataset","teacher_disagreement_score":0.01595765,"threshold_uncertainty_score":0.9996833,"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."}}