Rocky
152 lines
6.3 KiB
JSON
152 lines
6.3 KiB
JSON
{
|
|
"classifiers" : [
|
|
{
|
|
"candidates" : [
|
|
{
|
|
"acceptedForAutomaticRouting" : true,
|
|
"algorithm" : "maxEnt",
|
|
"balancedTrainingCount" : 5760,
|
|
"balancedValidationCount" : 1440,
|
|
"binaryByLanguage" : {
|
|
"en" : {
|
|
"accuracy" : 0.8903,
|
|
"f1" : 0.7311,
|
|
"falseNegative" : 152,
|
|
"falsePositive" : 1,
|
|
"precision" : 0.9952,
|
|
"recall" : 0.5778,
|
|
"total" : 1395,
|
|
"trueNegative" : 1034,
|
|
"truePositive" : 208
|
|
},
|
|
"zh-Hans" : {
|
|
"accuracy" : 0.9527,
|
|
"f1" : 0.8991,
|
|
"falseNegative" : 66,
|
|
"falsePositive" : 0,
|
|
"precision" : 1,
|
|
"recall" : 0.8167,
|
|
"total" : 1395,
|
|
"trueNegative" : 1035,
|
|
"truePositive" : 294
|
|
}
|
|
},
|
|
"confidenceThresholdsByLanguage" : {
|
|
"en" : 0.91,
|
|
"zh-Hans" : 0.9
|
|
},
|
|
"goldenBinary" : {
|
|
"accuracy" : 0.9434,
|
|
"f1" : 0.7273,
|
|
"falseNegative" : 12,
|
|
"falsePositive" : 0,
|
|
"precision" : 1,
|
|
"recall" : 0.5714,
|
|
"total" : 212,
|
|
"trueNegative" : 184,
|
|
"truePositive" : 16
|
|
},
|
|
"goldenBinaryByLanguage" : {
|
|
"en" : {
|
|
"accuracy" : 0.9528,
|
|
"f1" : 0.7826,
|
|
"falseNegative" : 5,
|
|
"falsePositive" : 0,
|
|
"precision" : 1,
|
|
"recall" : 0.6429,
|
|
"total" : 106,
|
|
"trueNegative" : 92,
|
|
"truePositive" : 9
|
|
},
|
|
"zh-Hans" : {
|
|
"accuracy" : 0.934,
|
|
"f1" : 0.6667,
|
|
"falseNegative" : 7,
|
|
"falsePositive" : 0,
|
|
"precision" : 1,
|
|
"recall" : 0.5,
|
|
"total" : 106,
|
|
"trueNegative" : 92,
|
|
"truePositive" : 7
|
|
}
|
|
},
|
|
"goldenFalseNegativeExamples" : [
|
|
"[en] Do not forget to check whether the refund arrives on Friday. (score=0.899)",
|
|
"[en] After the meeting, summarize the three decisions in the project channel. (score=0.7019)",
|
|
"[en] When approval arrives, contact the applicant and explain the reason. (score=0.4227)",
|
|
"[en] Follow up with the vendor about the delivery date next Monday. (score=0.9017)",
|
|
"[en] Next action: create the release tag after the tests pass. (score=0.5645)",
|
|
"[zh-Hans] 小王负责整理会议纪要，今天发到项目群。 (score=0.5646)",
|
|
"[zh-Hans] 方便帮我约一下周三下午的会议室吗？ (score=0.0988)",
|
|
"[zh-Hans] 这些材料可以在月底前准备好吗？ (score=0.575)",
|
|
"[zh-Hans] 别忘了周五检查退款有没有到账。 (score=0.6978)",
|
|
"[zh-Hans] 会后需要整理三个决定并发到项目群。 (score=0.129)",
|
|
"[zh-Hans] 等审批结果出来，请联系申请人说明原因。 (score=0.7829)",
|
|
"[zh-Hans] 下周一跟进供应商的交货日期。 (score=0.0068)"
|
|
],
|
|
"goldenFalsePositiveExamples" : [
|
|
|
|
],
|
|
"modelBytes" : 27275,
|
|
"runtimeAssetIndependent" : true,
|
|
"testBinary" : {
|
|
"accuracy" : 0.9215,
|
|
"f1" : 0.8209,
|
|
"falseNegative" : 218,
|
|
"falsePositive" : 1,
|
|
"precision" : 0.998,
|
|
"recall" : 0.6972,
|
|
"total" : 2790,
|
|
"trueNegative" : 2069,
|
|
"truePositive" : 502
|
|
},
|
|
"testFalseNegativeExamples" : [
|
|
"[en] Also, Reminder: call the client back before the next meeting. (score=0.7217)",
|
|
"[en] One more thing: Reminder: book the follow-up visit by Friday. (score=0.6472)",
|
|
"[en] Reminder: check the invoice status by the end of the month. (score=0.8201)",
|
|
"[en] Reminder: call the client back by the end of the month. (score=0.8283)",
|
|
"[en] A quick note: Reminder: call the client back before the end of today. (score=0.7477)",
|
|
"[en] One more thing: Reminder: create the release tag before 3 PM. (score=0.6731)",
|
|
"[en] Reminder: contact the applicant by the end of the month. (score=0.7488)",
|
|
"[en] Also, Make review the approval result the next action and do it by the end of the month. (score=0.8723)",
|
|
"[en] A quick note: Make check the vendor delivery date the next action and do it by the end of the month. (score=0.8494)",
|
|
"[en] Reminder: contact the applicant after you receive this message. (score=0.754)",
|
|
"[en] A quick note: Reminder: verify the refund status before the next meeting. (score=0.7248)",
|
|
"[en] A quick note: Reminder: record my temperature tomorrow morning. (score=0.5867)"
|
|
],
|
|
"testFalsePositiveExamples" : [
|
|
"[en] Todo visible only to me: prepare the presentation deck. (score=0.9134)"
|
|
],
|
|
"threshold" : 0.9,
|
|
"validationBinary" : {
|
|
"accuracy" : 0.9258,
|
|
"f1" : 0.8358,
|
|
"falseNegative" : 193,
|
|
"falsePositive" : 14,
|
|
"precision" : 0.9741,
|
|
"recall" : 0.7319,
|
|
"total" : 2790,
|
|
"trueNegative" : 2056,
|
|
"truePositive" : 527
|
|
}
|
|
}
|
|
],
|
|
"id" : "task",
|
|
"labels" : [
|
|
"notTask",
|
|
"task"
|
|
],
|
|
"positiveLabel" : "task",
|
|
"selectedAlgorithm" : "maxEnt",
|
|
"selectedModelFile" : "TaskIntentClassifier.mlmodel"
|
|
}
|
|
],
|
|
"corpusCount" : 16572,
|
|
"corpusPath" : "ModelTraining/ClipboardSemantics/clipboard_semantic_corpus.jsonl",
|
|
"generatedAt" : "2026-08-27T01:11:10Z",
|
|
"goldenCount" : 212,
|
|
"selectionPolicy" : "Validation only: global and per-language binary thresholds require precision >= 0.97, then maximize recall; languages with fewer than 20 examples per class fall back to the global threshold. sentiment prioritizes macro-F1. Automatic routing also requires a self-contained maxEnt model because BERT embedding assets are not guaranteed in extensions. Test and golden data gate deployment but never tune model weights.",
|
|
"testCount" : 2790,
|
|
"trainingCount" : 10780,
|
|
"validationCount" : 2790
|
|
} |