{"id":"W3087820244","doi":"10.1145/3395027.3419587","title":"A Framework to Evaluate Webpage Segment Recognizers","year":2020,"lang":"en","type":"article","venue":"","topic":"Web Data Mining and Analysis","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University","funders":"","keywords":"Computer science; Web page; Information retrieval; Search engine; World Wide Web; 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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0002280769,0.00008585898,0.0001259819,0.00004622621,0.00005661011,0.0001541078,0.0007578215,0.00003021166,0.0002994852],"category_scores_gemma":[0.0002036621,0.00007100769,0.00005766344,0.0007136581,0.000007321824,0.0001595628,0.0003655483,0.00009231532,0.002246612],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001363496,"about_ca_system_score_gemma":0.00003317028,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001669053,"about_ca_topic_score_gemma":0.000002080182,"domain_scores_codex":[0.9989776,0.00004619981,0.0001369257,0.0003995527,0.0002467019,0.0001930157],"domain_scores_gemma":[0.9991997,0.00005989273,0.0000252704,0.0004203903,0.00003488883,0.0002598072],"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":[0.0000197407,0.00009194426,0.001125034,0.00002318553,0.000207513,0.00007511284,0.009002476,0.001297118,0.00243752,0.02689029,0.4213894,0.5374407],"study_design_scores_gemma":[0.0007277604,0.0007310946,0.001018972,0.0001107059,0.0001048159,0.000007523934,0.0007799949,0.7635728,0.011087,0.006064787,0.2146424,0.001152198],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.005996353,0.00001559775,0.9463523,0.04057115,0.00009626717,0.00005856993,0.000002858218,0.000217313,0.006689577],"genre_scores_gemma":[0.3033847,0.000005631942,0.6706897,0.02481639,0.0001073537,0.00001048651,0.000003302305,0.000006561293,0.0009758432],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7622756,"threshold_uncertainty_score":0.9985303,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05614677183258542,"score_gpt":0.2995525377857658,"score_spread":0.2434057659531804,"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."}}