{"id":"W1921587136","doi":"10.1007/s10579-015-9318-3","title":"Cross level semantic similarity: an evaluation framework for universal measures of similarity","year":2015,"lang":"en","type":"article","venue":"Language Resources and Evaluation","topic":"Topic Modeling","field":"Computer Science","cited_by":14,"is_retracted":false,"has_abstract":false,"ca_institutions":"McGill University","funders":"European Research Council","keywords":"Computer science; Natural language processing; Similarity (geometry); Semantic similarity; Sentence; Task (project management); Artificial intelligence; Word (group theory); SemEval; Paragraph; Process (computing); Meaning (existential); WordNet; Information retrieval; Linguistics; Psychology; World Wide Web","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":[0.02412921,0.001391183,0.00207106,0.01212362,0.001430447,0.005369069,0.002473444,0.002148517,0.002995464],"category_scores_gemma":[0.06528036,0.0005231361,0.001974848,0.007821052,0.001956835,0.01136366,0.005723787,0.002126443,0.0008729197],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001951346,"about_ca_system_score_gemma":0.001957381,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001939076,"about_ca_topic_score_gemma":0.001862738,"domain_scores_codex":[0.9749388,0.01090835,0.002760674,0.002648959,0.008026661,0.0007165587],"domain_scores_gemma":[0.9645081,0.01954816,0.002398135,0.006381042,0.005970902,0.001193555],"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":[0.001907163,0.0009123082,0.02824858,0.001678404,0.001776341,0.0002376801,0.002643757,0.02845666,0.01970136,0.1872264,0.01001344,0.7171978],"study_design_scores_gemma":[0.0002019675,0.002026813,0.02248336,0.0005611283,0.001246642,0.001179397,0.001584848,0.5581739,0.02920008,0.3625684,0.02039169,0.000381724],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02168853,0.001003108,0.9720469,0.0001430849,0.00008880923,0.0004102549,0.0006332746,0.001345997,0.002640011],"genre_scores_gemma":[0.4087933,0.0005545408,0.5853214,0.0001463601,0.0001934645,0.001135799,0.001917916,0.0005842705,0.001353115],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02412921,"threshold_uncertainty_score":0.127609,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.212328201095616,"score_gpt":0.3869887300338786,"score_spread":0.1746605289382627,"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."}}