File size: 27,238 Bytes
edf1149 3b1e6c7 edf1149 3b1e6c7 edf1149 3b1e6c7 edf1149 3b1e6c7 edf1149 3b1e6c7 edf1149 3b1e6c7 edf1149 3b1e6c7 edf1149 3b1e6c7 edf1149 3b1e6c7 edf1149 3b1e6c7 edf1149 3b1e6c7 edf1149 3b1e6c7 edf1149 3b1e6c7 edf1149 3b1e6c7 edf1149 3b1e6c7 edf1149 3b1e6c7 edf1149 3b1e6c7 edf1149 3b1e6c7 edf1149 3b1e6c7 edf1149 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 |
# DEPENDENCIES
from enum import Enum
from typing import Dict
from typing import Tuple
from dataclasses import dataclass
class Domain(Enum):
"""
Text domains for adaptive thresholding
"""
# Core domains
GENERAL = "general"
ACADEMIC = "academic"
CREATIVE = "creative"
AI_ML = "ai_ml"
SOFTWARE_DEV = "software_dev"
TECHNICAL_DOC = "technical_doc"
ENGINEERING = "engineering"
SCIENCE = "science"
BUSINESS = "business"
LEGAL = "legal"
MEDICAL = "medical"
JOURNALISM = "journalism"
MARKETING = "marketing"
SOCIAL_MEDIA = "social_media"
BLOG_PERSONAL = "blog_personal"
TUTORIAL = "tutorial"
class ConfidenceLevel(Enum):
"""
Confidence levels for classification
"""
VERY_LOW = "very_low"
LOW = "low"
MEDIUM = "medium"
HIGH = "high"
VERY_HIGH = "very_high"
@dataclass
class MetricThresholds:
"""
Thresholds for a single metric
"""
ai_threshold : float # Above this = likely AI
human_threshold : float # Below this = likely human
confidence_multiplier : float = 1.0
weight : float = 1.0
@dataclass
class DomainThresholds:
"""
Thresholds for 6 metrics in a specific domain
"""
domain : Domain
structural : MetricThresholds
perplexity : MetricThresholds
entropy : MetricThresholds
semantic_analysis : MetricThresholds
linguistic : MetricThresholds
multi_perturbation_stability : MetricThresholds
ensemble_threshold : float = 0.5
# ==================== DOMAIN-SPECIFIC THRESHOLDS ====================
# GENERAL (Default fallback)
DEFAULT_THRESHOLDS = DomainThresholds(domain = Domain.GENERAL,
structural = MetricThresholds(ai_threshold = 0.55, human_threshold = 0.45, weight = 0.20),
perplexity = MetricThresholds(ai_threshold = 0.52, human_threshold = 0.48, weight = 0.25),
entropy = MetricThresholds(ai_threshold = 0.48, human_threshold = 0.52, weight = 0.15),
semantic_analysis = MetricThresholds(ai_threshold = 0.55, human_threshold = 0.45, weight = 0.18),
linguistic = MetricThresholds(ai_threshold = 0.58, human_threshold = 0.42, weight = 0.12),
multi_perturbation_stability = MetricThresholds(ai_threshold = 0.60, human_threshold = 0.40, weight = 0.10),
ensemble_threshold = 0.40,
)
# ACADEMIC
ACADEMIC_THRESHOLDS = DomainThresholds(domain = Domain.ACADEMIC,
structural = MetricThresholds(ai_threshold = 0.58, human_threshold = 0.42, weight = 0.18),
perplexity = MetricThresholds(ai_threshold = 0.50, human_threshold = 0.45, weight = 0.26),
entropy = MetricThresholds(ai_threshold = 0.45, human_threshold = 0.50, weight = 0.14),
semantic_analysis = MetricThresholds(ai_threshold = 0.58, human_threshold = 0.42, weight = 0.20),
linguistic = MetricThresholds(ai_threshold = 0.62, human_threshold = 0.38, weight = 0.14),
multi_perturbation_stability = MetricThresholds(ai_threshold = 0.65, human_threshold = 0.35, weight = 0.08),
ensemble_threshold = 0.42,
)
# CREATIVE WRITING
CREATIVE_THRESHOLDS = DomainThresholds(domain = Domain.CREATIVE,
structural = MetricThresholds(ai_threshold = 0.52, human_threshold = 0.48, weight = 0.18),
perplexity = MetricThresholds(ai_threshold = 0.55, human_threshold = 0.50, weight = 0.22),
entropy = MetricThresholds(ai_threshold = 0.50, human_threshold = 0.55, weight = 0.16),
semantic_analysis = MetricThresholds(ai_threshold = 0.52, human_threshold = 0.48, weight = 0.20),
linguistic = MetricThresholds(ai_threshold = 0.55, human_threshold = 0.45, weight = 0.16),
multi_perturbation_stability = MetricThresholds(ai_threshold = 0.58, human_threshold = 0.42, weight = 0.08),
ensemble_threshold = 0.38,
)
# AI/ML/DATA SCIENCE
AI_ML_THRESHOLDS = DomainThresholds(domain = Domain.AI_ML,
structural = MetricThresholds(ai_threshold = 0.57, human_threshold = 0.43, weight = 0.18),
perplexity = MetricThresholds(ai_threshold = 0.51, human_threshold = 0.46, weight = 0.26),
entropy = MetricThresholds(ai_threshold = 0.47, human_threshold = 0.50, weight = 0.14),
semantic_analysis = MetricThresholds(ai_threshold = 0.57, human_threshold = 0.43, weight = 0.20),
linguistic = MetricThresholds(ai_threshold = 0.61, human_threshold = 0.39, weight = 0.14),
multi_perturbation_stability = MetricThresholds(ai_threshold = 0.64, human_threshold = 0.36, weight = 0.08),
ensemble_threshold = 0.41,
)
# SOFTWARE DEVELOPMENT
SOFTWARE_DEV_THRESHOLDS = DomainThresholds(domain = Domain.SOFTWARE_DEV,
structural = MetricThresholds(ai_threshold = 0.58, human_threshold = 0.42, weight = 0.17),
perplexity = MetricThresholds(ai_threshold = 0.50, human_threshold = 0.45, weight = 0.27),
entropy = MetricThresholds(ai_threshold = 0.46, human_threshold = 0.50, weight = 0.14),
semantic_analysis = MetricThresholds(ai_threshold = 0.58, human_threshold = 0.42, weight = 0.20),
linguistic = MetricThresholds(ai_threshold = 0.60, human_threshold = 0.40, weight = 0.14),
multi_perturbation_stability = MetricThresholds(ai_threshold = 0.63, human_threshold = 0.37, weight = 0.08),
ensemble_threshold = 0.41,
)
# TECHNICAL DOCUMENTATION
TECHNICAL_DOC_THRESHOLDS = DomainThresholds(domain = Domain.TECHNICAL_DOC,
structural = MetricThresholds(ai_threshold = 0.59, human_threshold = 0.41, weight = 0.18),
perplexity = MetricThresholds(ai_threshold = 0.49, human_threshold = 0.44, weight = 0.27),
entropy = MetricThresholds(ai_threshold = 0.45, human_threshold = 0.49, weight = 0.13),
semantic_analysis = MetricThresholds(ai_threshold = 0.59, human_threshold = 0.41, weight = 0.20),
linguistic = MetricThresholds(ai_threshold = 0.62, human_threshold = 0.38, weight = 0.14),
multi_perturbation_stability = MetricThresholds(ai_threshold = 0.65, human_threshold = 0.35, weight = 0.08),
ensemble_threshold = 0.42,
)
# ENGINEERING
ENGINEERING_THRESHOLDS = DomainThresholds(domain = Domain.ENGINEERING,
structural = MetricThresholds(ai_threshold = 0.58, human_threshold = 0.42, weight = 0.18),
perplexity = MetricThresholds(ai_threshold = 0.50, human_threshold = 0.45, weight = 0.26),
entropy = MetricThresholds(ai_threshold = 0.46, human_threshold = 0.50, weight = 0.14),
semantic_analysis = MetricThresholds(ai_threshold = 0.58, human_threshold = 0.42, weight = 0.20),
linguistic = MetricThresholds(ai_threshold = 0.61, human_threshold = 0.39, weight = 0.14),
multi_perturbation_stability = MetricThresholds(ai_threshold = 0.64, human_threshold = 0.36, weight = 0.08),
ensemble_threshold = 0.41,
)
# SCIENCE (Physics, Chemistry, Biology)
SCIENCE_THRESHOLDS = DomainThresholds(domain = Domain.SCIENCE,
structural = MetricThresholds(ai_threshold = 0.58, human_threshold = 0.42, weight = 0.18),
perplexity = MetricThresholds(ai_threshold = 0.51, human_threshold = 0.46, weight = 0.26),
entropy = MetricThresholds(ai_threshold = 0.46, human_threshold = 0.50, weight = 0.14),
semantic_analysis = MetricThresholds(ai_threshold = 0.58, human_threshold = 0.42, weight = 0.20),
linguistic = MetricThresholds(ai_threshold = 0.62, human_threshold = 0.38, weight = 0.14),
multi_perturbation_stability = MetricThresholds(ai_threshold = 0.64, human_threshold = 0.36, weight = 0.08),
ensemble_threshold = 0.42,
)
# BUSINESS
BUSINESS_THRESHOLDS = DomainThresholds(domain = Domain.BUSINESS,
structural = MetricThresholds(ai_threshold = 0.56, human_threshold = 0.44, weight = 0.18),
perplexity = MetricThresholds(ai_threshold = 0.52, human_threshold = 0.48, weight = 0.24),
entropy = MetricThresholds(ai_threshold = 0.48, human_threshold = 0.52, weight = 0.15),
semantic_analysis = MetricThresholds(ai_threshold = 0.56, human_threshold = 0.44, weight = 0.19),
linguistic = MetricThresholds(ai_threshold = 0.60, human_threshold = 0.40, weight = 0.15),
multi_perturbation_stability = MetricThresholds(ai_threshold = 0.62, human_threshold = 0.38, weight = 0.09),
ensemble_threshold = 0.40,
)
# LEGAL
LEGAL_THRESHOLDS = DomainThresholds(domain = Domain.LEGAL,
structural = MetricThresholds(ai_threshold = 0.60, human_threshold = 0.40, weight = 0.17),
perplexity = MetricThresholds(ai_threshold = 0.50, human_threshold = 0.44, weight = 0.27),
entropy = MetricThresholds(ai_threshold = 0.44, human_threshold = 0.48, weight = 0.13),
semantic_analysis = MetricThresholds(ai_threshold = 0.60, human_threshold = 0.40, weight = 0.20),
linguistic = MetricThresholds(ai_threshold = 0.63, human_threshold = 0.37, weight = 0.15),
multi_perturbation_stability = MetricThresholds(ai_threshold = 0.66, human_threshold = 0.34, weight = 0.08),
ensemble_threshold = 0.43,
)
# MEDICAL
MEDICAL_THRESHOLDS = DomainThresholds(domain = Domain.MEDICAL,
structural = MetricThresholds(ai_threshold = 0.59, human_threshold = 0.41, weight = 0.17),
perplexity = MetricThresholds(ai_threshold = 0.50, human_threshold = 0.45, weight = 0.27),
entropy = MetricThresholds(ai_threshold = 0.45, human_threshold = 0.49, weight = 0.13),
semantic_analysis = MetricThresholds(ai_threshold = 0.59, human_threshold = 0.41, weight = 0.20),
linguistic = MetricThresholds(ai_threshold = 0.62, human_threshold = 0.38, weight = 0.15),
multi_perturbation_stability = MetricThresholds(ai_threshold = 0.65, human_threshold = 0.35, weight = 0.08),
ensemble_threshold = 0.43,
)
# JOURNALISM
JOURNALISM_THRESHOLDS = DomainThresholds(domain = Domain.JOURNALISM,
structural = MetricThresholds(ai_threshold = 0.56, human_threshold = 0.44, weight = 0.18),
perplexity = MetricThresholds(ai_threshold = 0.52, human_threshold = 0.48, weight = 0.24),
entropy = MetricThresholds(ai_threshold = 0.48, human_threshold = 0.52, weight = 0.15),
semantic_analysis = MetricThresholds(ai_threshold = 0.56, human_threshold = 0.44, weight = 0.20),
linguistic = MetricThresholds(ai_threshold = 0.58, human_threshold = 0.42, weight = 0.15),
multi_perturbation_stability = MetricThresholds(ai_threshold = 0.62, human_threshold = 0.38, weight = 0.08),
ensemble_threshold = 0.40,
)
# MARKETING
MARKETING_THRESHOLDS = DomainThresholds(domain = Domain.MARKETING,
structural = MetricThresholds(ai_threshold = 0.54, human_threshold = 0.46, weight = 0.19),
perplexity = MetricThresholds(ai_threshold = 0.53, human_threshold = 0.49, weight = 0.23),
entropy = MetricThresholds(ai_threshold = 0.49, human_threshold = 0.53, weight = 0.15),
semantic_analysis = MetricThresholds(ai_threshold = 0.54, human_threshold = 0.46, weight = 0.19),
linguistic = MetricThresholds(ai_threshold = 0.57, human_threshold = 0.43, weight = 0.16),
multi_perturbation_stability = MetricThresholds(ai_threshold = 0.61, human_threshold = 0.39, weight = 0.08),
ensemble_threshold = 0.39,
)
# SOCIAL MEDIA
SOCIAL_MEDIA_THRESHOLDS = DomainThresholds(domain = Domain.SOCIAL_MEDIA,
structural = MetricThresholds(ai_threshold = 0.52, human_threshold = 0.48, weight = 0.18),
perplexity = MetricThresholds(ai_threshold = 0.54, human_threshold = 0.50, weight = 0.20),
entropy = MetricThresholds(ai_threshold = 0.50, human_threshold = 0.54, weight = 0.17),
semantic_analysis = MetricThresholds(ai_threshold = 0.52, human_threshold = 0.48, weight = 0.18),
linguistic = MetricThresholds(ai_threshold = 0.55, human_threshold = 0.45, weight = 0.18),
multi_perturbation_stability = MetricThresholds(ai_threshold = 0.60, human_threshold = 0.40, weight = 0.09),
ensemble_threshold = 0.36,
)
# PERSONAL BLOG
BLOG_PERSONAL_THRESHOLDS = DomainThresholds(domain = Domain.BLOG_PERSONAL,
structural = MetricThresholds(ai_threshold = 0.53, human_threshold = 0.47, weight = 0.19),
perplexity = MetricThresholds(ai_threshold = 0.54, human_threshold = 0.50, weight = 0.22),
entropy = MetricThresholds(ai_threshold = 0.50, human_threshold = 0.54, weight = 0.16),
semantic_analysis = MetricThresholds(ai_threshold = 0.53, human_threshold = 0.47, weight = 0.19),
linguistic = MetricThresholds(ai_threshold = 0.56, human_threshold = 0.44, weight = 0.16),
multi_perturbation_stability = MetricThresholds(ai_threshold = 0.59, human_threshold = 0.41, weight = 0.08),
ensemble_threshold = 0.38,
)
# TUTORIAL/HOW-TO
TUTORIAL_THRESHOLDS = DomainThresholds(domain = Domain.TUTORIAL,
structural = MetricThresholds(ai_threshold = 0.56, human_threshold = 0.44, weight = 0.18),
perplexity = MetricThresholds(ai_threshold = 0.52, human_threshold = 0.48, weight = 0.25),
entropy = MetricThresholds(ai_threshold = 0.48, human_threshold = 0.52, weight = 0.15),
semantic_analysis = MetricThresholds(ai_threshold = 0.56, human_threshold = 0.44, weight = 0.19),
linguistic = MetricThresholds(ai_threshold = 0.59, human_threshold = 0.41, weight = 0.15),
multi_perturbation_stability = MetricThresholds(ai_threshold = 0.62, human_threshold = 0.38, weight = 0.08),
ensemble_threshold = 0.40,
)
# THRESHOLD REGISTRY
THRESHOLD_REGISTRY: Dict[Domain, DomainThresholds] = {Domain.GENERAL : DEFAULT_THRESHOLDS,
Domain.ACADEMIC : ACADEMIC_THRESHOLDS,
Domain.CREATIVE : CREATIVE_THRESHOLDS,
Domain.AI_ML : AI_ML_THRESHOLDS,
Domain.SOFTWARE_DEV : SOFTWARE_DEV_THRESHOLDS,
Domain.TECHNICAL_DOC : TECHNICAL_DOC_THRESHOLDS,
Domain.ENGINEERING : ENGINEERING_THRESHOLDS,
Domain.SCIENCE : SCIENCE_THRESHOLDS,
Domain.BUSINESS : BUSINESS_THRESHOLDS,
Domain.LEGAL : LEGAL_THRESHOLDS,
Domain.MEDICAL : MEDICAL_THRESHOLDS,
Domain.JOURNALISM : JOURNALISM_THRESHOLDS,
Domain.MARKETING : MARKETING_THRESHOLDS,
Domain.SOCIAL_MEDIA : SOCIAL_MEDIA_THRESHOLDS,
Domain.BLOG_PERSONAL : BLOG_PERSONAL_THRESHOLDS,
Domain.TUTORIAL : TUTORIAL_THRESHOLDS,
}
# CONFIDENCE LEVEL RANGES
CONFIDENCE_RANGES: Dict[ConfidenceLevel, Tuple[float, float]] = {ConfidenceLevel.VERY_LOW : (0.0, 0.3),
ConfidenceLevel.LOW : (0.3, 0.5),
ConfidenceLevel.MEDIUM : (0.5, 0.7),
ConfidenceLevel.HIGH : (0.7, 0.85),
ConfidenceLevel.VERY_HIGH : (0.85, 1.0),
}
# HELPER FUNCTIONS
def get_threshold_for_domain(domain: Domain) -> DomainThresholds:
"""
Get thresholds for a specific domain
"""
return THRESHOLD_REGISTRY.get(domain, DEFAULT_THRESHOLDS)
def get_confidence_level(score: float) -> ConfidenceLevel:
"""
Determine confidence level based on score
"""
for level, (min_val, max_val) in CONFIDENCE_RANGES.items():
if (min_val <= score < max_val):
return level
return ConfidenceLevel.VERY_HIGH
def adjust_threshold_by_confidence(threshold: float, confidence: float, conservative: bool = True) -> float:
"""
Adjust threshold based on confidence level
"""
if conservative:
adjustment = (1 - confidence) * 0.1
adjusted_threshold = threshold + adjustment
return adjusted_threshold
else:
adjustment = confidence * 0.05
adjusted_threshold = threshold - adjustment
return adjusted_threshold
def interpolate_thresholds(domain1: Domain, domain2: Domain, weight1: float = 0.5) -> DomainThresholds:
"""
Interpolate between two domain thresholds
"""
thresh1 = get_threshold_for_domain(domain = domain1)
thresh2 = get_threshold_for_domain(domain = domain2)
weight2 = 1 - weight1
def interpolate_metric(m1: MetricThresholds, m2: MetricThresholds) -> MetricThresholds:
return MetricThresholds(ai_threshold = m1.ai_threshold * weight1 + m2.ai_threshold * weight2,
human_threshold = m1.human_threshold * weight1 + m2.human_threshold * weight2,
weight = m1.weight * weight1 + m2.weight * weight2,
)
return DomainThresholds(domain = domain1,
structural = interpolate_metric(thresh1.structural, thresh2.structural),
perplexity = interpolate_metric(thresh1.perplexity, thresh2.perplexity),
entropy = interpolate_metric(thresh1.entropy, thresh2.entropy),
semantic_analysis = interpolate_metric(thresh1.semantic_analysis, thresh2.semantic_analysis),
linguistic = interpolate_metric(thresh1.linguistic, thresh2.linguistic),
multi_perturbation_stability = interpolate_metric(thresh1.multi_perturbation_stability, thresh2.multi_perturbation_stability),
ensemble_threshold = thresh1.ensemble_threshold * weight1 + thresh2.ensemble_threshold * weight2,
)
def get_active_metric_weights(domain: Domain, enabled_metrics: Dict[str, bool]) -> Dict[str, float]:
"""
Get weights for enabled metrics, normalized to sum to 1.0
"""
thresholds = get_threshold_for_domain(domain = domain)
metric_mapping = {"structural" : thresholds.structural,
"perplexity" : thresholds.perplexity,
"entropy" : thresholds.entropy,
"semantic_analysis" : thresholds.semantic_analysis,
"linguistic" : thresholds.linguistic,
"multi_perturbation_stability" : thresholds.multi_perturbation_stability,
}
active_weights = dict()
for metric_name, threshold_obj in metric_mapping.items():
if enabled_metrics.get(metric_name, False):
active_weights[metric_name] = threshold_obj.weight
# Normalize
total_weight = sum(active_weights.values())
if (total_weight > 0):
active_weights = {name: weight / total_weight for name, weight in active_weights.items()}
return active_weights
# Export
__all__ = ["Domain",
"ConfidenceLevel",
"MetricThresholds",
"DomainThresholds",
"CONFIDENCE_RANGES",
"DEFAULT_THRESHOLDS",
"THRESHOLD_REGISTRY",
"get_confidence_level",
"interpolate_thresholds",
"get_threshold_for_domain",
"get_active_metric_weights",
"adjust_threshold_by_confidence",
] |