Spaces:
Sleeping
Sleeping
File size: 64,557 Bytes
8205d6b 38d5865 c79f796 8205d6b 38d5865 8205d6b 38d5865 8205d6b c79f796 8205d6b c79f796 8205d6b c79f796 8205d6b 38d5865 8205d6b 38d5865 8205d6b 38d5865 8205d6b c79f796 8205d6b c79f796 8205d6b c79f796 8205d6b 38d5865 8205d6b 38d5865 8205d6b 38d5865 8205d6b 38d5865 8205d6b 38d5865 8205d6b 38d5865 8205d6b 38d5865 8205d6b 38d5865 8205d6b c79f796 8205d6b c79f796 8205d6b c79f796 8205d6b c79f796 38d5865 8205d6b c79f796 38d5865 8205d6b c79f796 8205d6b c79f796 8205d6b c79f796 38d5865 c79f796 8205d6b 38d5865 8205d6b c79f796 8205d6b 38d5865 c79f796 8205d6b c79f796 38d5865 c79f796 8205d6b 38d5865 8205d6b |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 |
# DEPENDENCIES
import re
import numpy as np
from enum import Enum
from typing import Any
from typing import Dict
from typing import List
from typing import Tuple
from loguru import logger
from typing import Optional
from dataclasses import dataclass
from config.threshold_config import Domain
from metrics.base_metric import MetricResult
from processors.text_processor import ProcessedText
class AIModel(Enum):
"""
Supported AI models for attribution - ALIGNED WITH DOCUMENTATION
"""
GPT_3_5 = "gpt-3.5-turbo"
GPT_4 = "gpt-4"
GPT_4_TURBO = "gpt-4-turbo"
GPT_4o = "gpt-4o"
CLAUDE_3_OPUS = "claude-3-opus"
CLAUDE_3_SONNET = "claude-3-sonnet"
CLAUDE_3_HAIKU = "claude-3-haiku"
GEMINI_PRO = "gemini-pro"
GEMINI_ULTRA = "gemini-ultra"
GEMINI_FLASH = "gemini-flash"
LLAMA_2 = "llama-2"
LLAMA_3 = "llama-3"
MISTRAL = "mistral"
MIXTRAL = "mixtral"
DEEPSEEK_CHAT = "deepseek-chat"
DEEPSEEK_CODER = "deepseek-coder"
HUMAN = "human"
UNKNOWN = "unknown"
@dataclass
class AttributionResult:
"""
Result of AI model attribution
"""
predicted_model : AIModel
confidence : float
model_probabilities : Dict[str, float]
reasoning : List[str]
fingerprint_matches : Dict[str, int]
domain_used : Domain
metric_contributions: Dict[str, float]
def to_dict(self) -> Dict[str, Any]:
"""
Convert to dictionary
"""
return {"predicted_model" : self.predicted_model.value,
"confidence" : round(self.confidence, 4),
"model_probabilities" : {model: round(prob, 4) for model, prob in self.model_probabilities.items()},
"reasoning" : self.reasoning,
"fingerprint_matches" : self.fingerprint_matches,
"domain_used" : self.domain_used.value,
"metric_contributions": {metric: round(contrib, 4) for metric, contrib in self.metric_contributions.items()},
}
class ModelAttributor:
"""
Model attribution
FEATURES:
- Domain-aware calibration
- 6-metric ensemble integration
- Confidence-weighted aggregation
- Explainable reasoning
"""
# Metric weights from technical specification
METRIC_WEIGHTS = {"perplexity" : 0.25,
"structural" : 0.15,
"semantic_analysis" : 0.15,
"entropy" : 0.20,
"linguistic" : 0.15,
"multi_perturbation_stability" : 0.10,
}
# Domain-aware model patterns for ALL 16 DOMAINS
DOMAIN_MODEL_PREFERENCES = {Domain.GENERAL : [AIModel.GPT_4, AIModel.CLAUDE_3_SONNET, AIModel.GEMINI_PRO, AIModel.GPT_3_5],
Domain.ACADEMIC : [AIModel.GPT_4, AIModel.CLAUDE_3_OPUS, AIModel.GEMINI_ULTRA, AIModel.GPT_4_TURBO],
Domain.TECHNICAL_DOC : [AIModel.GPT_4_TURBO, AIModel.CLAUDE_3_SONNET, AIModel.LLAMA_3, AIModel.GPT_4],
Domain.AI_ML : [AIModel.GPT_4_TURBO, AIModel.GPT_4, AIModel.CLAUDE_3_OPUS, AIModel.DEEPSEEK_CODER],
Domain.SOFTWARE_DEV : [AIModel.GPT_4_TURBO, AIModel.DEEPSEEK_CODER, AIModel.CLAUDE_3_SONNET, AIModel.GPT_4],
Domain.ENGINEERING : [AIModel.GPT_4, AIModel.CLAUDE_3_OPUS, AIModel.GPT_4_TURBO, AIModel.LLAMA_3],
Domain.SCIENCE : [AIModel.GPT_4, AIModel.CLAUDE_3_OPUS, AIModel.GEMINI_ULTRA, AIModel.GPT_4_TURBO],
Domain.BUSINESS : [AIModel.GPT_4, AIModel.CLAUDE_3_SONNET, AIModel.GEMINI_PRO, AIModel.GPT_3_5],
Domain.LEGAL : [AIModel.GPT_4, AIModel.CLAUDE_3_OPUS, AIModel.GPT_4_TURBO, AIModel.CLAUDE_3_SONNET],
Domain.MEDICAL : [AIModel.GPT_4, AIModel.CLAUDE_3_OPUS, AIModel.GEMINI_ULTRA, AIModel.GPT_4_TURBO],
Domain.JOURNALISM : [AIModel.GPT_4, AIModel.CLAUDE_3_SONNET, AIModel.GEMINI_PRO, AIModel.GPT_3_5],
Domain.CREATIVE : [AIModel.CLAUDE_3_OPUS, AIModel.GPT_4, AIModel.GEMINI_PRO, AIModel.CLAUDE_3_SONNET],
Domain.MARKETING : [AIModel.GPT_4, AIModel.CLAUDE_3_SONNET, AIModel.GEMINI_PRO, AIModel.GPT_3_5],
Domain.SOCIAL_MEDIA : [AIModel.GPT_3_5, AIModel.GEMINI_PRO, AIModel.DEEPSEEK_CHAT, AIModel.LLAMA_3],
Domain.BLOG_PERSONAL : [AIModel.CLAUDE_3_SONNET, AIModel.GPT_4, AIModel.GEMINI_PRO, AIModel.GPT_3_5],
Domain.TUTORIAL : [AIModel.GPT_4, AIModel.CLAUDE_3_SONNET, AIModel.GEMINI_PRO, AIModel.GPT_4_TURBO],
}
# Model-specific fingerprints with comprehensive patterns
MODEL_FINGERPRINTS = {AIModel.GPT_3_5 : {"phrases" : ["as an ai language model",
"i don't have personal opinions",
"it's important to note that",
"it's worth noting that",
"keep in mind that",
"bear in mind that",
"i should point out",
"it's also important to",
"additionally, it's worth",
"furthermore, it should be",
"i cannot provide",
"i'm unable to",
"i don't have the ability",
"based on the information",
"according to the context",
],
"sentence_starters" : ["however,",
"additionally,",
"furthermore,",
"moreover,",
"in conclusion,",
"therefore,",
"consequently,",
"as a result,",
"in summary,",
"ultimately,",
],
"structural_patterns" : ["firstly",
"secondly",
"thirdly",
"on one hand",
"on the other hand",
"in terms of",
"with regard to",
],
"punctuation_patterns" : {"em_dash_frequency" : (0.01, 0.03),
"semicolon_frequency" : (0.005, 0.015),
"parentheses_frequency" : (0.01, 0.04),
},
"style_markers" : {"avg_sentence_length" : (18, 25),
"transition_word_density" : (0.08, 0.15),
"formality_score" : (0.7, 0.9),
"hedging_language" : (0.05, 0.12),
}
},
AIModel.GPT_4 : {"phrases" : ["it's important to note that",
"it's worth mentioning that",
"to clarify this point",
"in other words,",
"that being said,",
"in essence,",
"fundamentally,",
"at its core,",
"from a broader perspective",
"when considering",
"this suggests that",
"this implies that",
"it follows that",
"consequently,",
"accordingly,",
],
"sentence_starters" : ["interestingly,",
"notably,",
"crucially,",
"essentially,",
"ultimately,",
"significantly,",
"importantly,",
"remarkably,",
"surprisingly,",
],
"structural_patterns" : ["in light of",
"with respect to",
"pertaining to",
"as evidenced by",
"as indicated by",
"as suggested by",
],
"punctuation_patterns" : {"em_dash_frequency" : (0.02, 0.05),
"colon_frequency" : (0.01, 0.03),
"semicolon_frequency" : (0.01, 0.02),
},
"style_markers" : {"avg_sentence_length" : (20, 28),
"vocabulary_sophistication" : (0.7, 0.9),
"conceptual_density" : (0.6, 0.85),
"analytical_depth" : (0.65, 0.9),
}
},
AIModel.CLAUDE_3_OPUS : {"phrases" : ["i'd be glad to",
"i'm happy to help",
"let me explain this",
"to clarify this further",
"in this context,",
"from this perspective,",
"building on that point",
"expanding on this idea",
"delving deeper into",
"to elaborate further",
"it's worth considering",
"this raises the question",
"this highlights the importance",
"this underscores the need",
],
"sentence_starters" : ["certainly,",
"indeed,",
"particularly,",
"specifically,",
"notably,",
"importantly,",
"interestingly,",
"crucially,",
],
"structural_patterns" : ["in other words",
"to put it differently",
"that is to say",
"for instance",
"for example",
"as an illustration",
],
"punctuation_patterns" : {"em_dash_frequency" : (0.015, 0.04),
"parenthetical_usage" : (0.02, 0.06),
"colon_frequency" : (0.008, 0.025),
},
"style_markers" : {"avg_sentence_length" : (17, 24),
"nuanced_language" : (0.6, 0.85),
"explanatory_depth" : (0.7, 0.95),
"conceptual_clarity" : (0.65, 0.9),
}
},
AIModel.GEMINI_PRO : {"phrases" : ["here's what you need to know",
"here's how it works",
"let's explore this",
"let's look at this",
"consider this example",
"think of it this way",
"imagine if you will",
"picture this scenario",
"to break it down",
"in simple terms",
"put simply,",
"basically,",
"the key point is",
"the main idea here",
],
"sentence_starters" : ["now,",
"so,",
"well,",
"basically,",
"essentially,",
"actually,",
"technically,",
"practically,",
],
"structural_patterns" : ["on that note",
"speaking of which",
"by the way",
"as a side note",
"incidentally",
"in any case",
],
"punctuation_patterns" : {"exclamation_frequency" : (0.01, 0.03),
"question_frequency" : (0.02, 0.05),
"ellipsis_frequency" : (0.005, 0.02),
},
"style_markers" : {"avg_sentence_length" : (15, 22),
"conversational_tone" : (0.5, 0.8),
"accessibility_score" : (0.6, 0.9),
"engagement_level" : (0.55, 0.85),
}
},
AIModel.LLAMA_3 : {"phrases" : ["it's worth noting",
"it's important to understand",
"this means that",
"this indicates that",
"this shows that",
"this demonstrates that",
"based on this,",
"given this context",
"in this case,",
"for this reason",
"as such,",
"therefore,",
],
"sentence_starters" : ["first,",
"second,",
"third,",
"next,",
"then,",
"finally,",
"overall,",
"in general,",
],
"structural_patterns" : ["in addition",
"moreover",
"furthermore",
"however",
"nevertheless",
"nonetheless",
],
"punctuation_patterns" : {"comma_frequency" : (0.08, 0.15),
"period_frequency" : (0.06, 0.12),
"conjunction_frequency" : (0.05, 0.1),
},
"style_markers" : {"avg_sentence_length" : (16, 23),
"directness_score" : (0.6, 0.85),
"clarity_score" : (0.65, 0.9),
"structural_consistency" : (0.7, 0.95),
}
},
AIModel.DEEPSEEK_CHAT : {"phrases" : ["i understand you're asking",
"let me help you with that",
"i can assist you with",
"regarding your question",
"to answer your question",
"in response to your query",
"based on your request",
"as per your question",
"concerning your inquiry",
"with respect to your question",
"i'll do my best to",
"i'll try to help you",
"allow me to explain",
"let me break it down",
],
"sentence_starters" : ["well,",
"okay,",
"so,",
"now,",
"first,",
"actually,",
"specifically,",
"generally,",
],
"structural_patterns" : ["in other words",
"to put it simply",
"that is",
"for example",
"for instance",
"such as",
],
"punctuation_patterns" : {"comma_frequency" : (0.07, 0.14),
"period_frequency" : (0.05, 0.11),
"question_frequency" : (0.01, 0.04),
},
"style_markers" : {"avg_sentence_length" : (14, 21),
"helpfulness_tone" : (0.6, 0.9),
"explanatory_style" : (0.55, 0.85),
"user_focus" : (0.65, 0.95),
}
},
AIModel.MIXTRAL : {"phrases" : ["it should be noted that",
"it is important to recognize",
"this suggests that",
"this implies that",
"this indicates that",
"from this we can see",
"based on this analysis",
"considering these points",
"taking into account",
"in light of these factors",
],
"sentence_starters" : ["however,",
"moreover,",
"furthermore,",
"additionally,",
"conversely,",
"similarly,",
"likewise,",
],
"structural_patterns" : ["on the one hand",
"on the other hand",
"in contrast",
"by comparison",
"as opposed to",
"rather than",
],
"punctuation_patterns" : {"semicolon_frequency" : (0.008, 0.02),
"colon_frequency" : (0.006, 0.018),
"parentheses_frequency" : (0.012, 0.035),
},
"style_markers" : {"avg_sentence_length" : (19, 26),
"analytical_tone" : (0.65, 0.9),
"comparative_language" : (0.5, 0.8),
"balanced_perspective" : (0.6, 0.85),
}
}
}
def __init__(self):
"""
Initialize model attributor with domain awareness
"""
self.is_initialized = False
logger.info("ModelAttributor initialized with domain-aware calibration")
def initialize(self) -> bool:
"""
Initialize attribution system
"""
try:
self.is_initialized = True
logger.success("Model attribution system initialized with metric ensemble")
return True
except Exception as e:
logger.error(f"Failed to initialize attribution system: {repr(e)}")
return False
def attribute(self, text: str, processed_text: Optional[ProcessedText] = None, metric_results: Optional[Dict[str, MetricResult]] = None,
domain: Domain = Domain.GENERAL) -> AttributionResult:
"""
Attribute text to specific AI model with domain awareness
Arguments:
----------
text { str } : Input text
processed_text { ProcessedText } : Processed text metadata
metric_results { dict } : Results from 6 core metrics
domain { Domain } : Text domain for calibration
Returns:
--------
{ AttributionResult } : Attribution result with domain context
"""
try:
# Get domain-specific model preferences
domain_preferences = self.DOMAIN_MODEL_PREFERENCES.get(domain, [AIModel.GPT_4, AIModel.CLAUDE_3_SONNET])
# Fingerprint analysis
fingerprint_scores = self._calculate_fingerprint_scores(text = text,
domain = domain,
)
# Statistical pattern analysis
statistical_scores = self._analyze_statistical_patterns(text = text,
domain = domain,
)
# Metric-based attribution using all 6 metrics
metric_scores = self._analyze_metric_patterns(metric_results = metric_results, domain = domain) if metric_results else {}
# Ensemble Combination
combined_scores, metric_contributions = self._combine_attribution_scores(fingerprint_scores = fingerprint_scores,
statistical_scores = statistical_scores,
metric_scores = metric_scores,
domain = domain,
)
# Domain-aware prediction : Always show the actual highest probability model
predicted_model, confidence = self._make_domain_aware_prediction(combined_scores = combined_scores,
domain = domain,
domain_preferences = domain_preferences,
)
# Reasoning with domain context
reasoning = self._generate_detailed_reasoning(predicted_model = predicted_model,
confidence = confidence,
domain = domain,
metric_contributions = metric_contributions,
combined_scores = combined_scores,
)
return AttributionResult(predicted_model = predicted_model,
confidence = confidence,
model_probabilities = combined_scores,
reasoning = reasoning,
fingerprint_matches = self._get_top_fingerprints(fingerprint_scores),
domain_used = domain,
metric_contributions = metric_contributions,
)
except Exception as e:
logger.error(f"Error in model attribution: {repr(e)}")
return self._create_unknown_result(domain)
def _calculate_fingerprint_scores(self, text: str, domain: Domain) -> Dict[AIModel, float]:
"""
Calculate fingerprint match scores with domain calibration - for all domains
"""
scores = {model: 0.0 for model in AIModel if model not in [AIModel.HUMAN, AIModel.UNKNOWN]}
# Adjust sensitivity based on all domains
domain_sensitivity = {Domain.GENERAL : 1.00,
Domain.ACADEMIC : 1.20,
Domain.CREATIVE : 0.90,
Domain.AI_ML : 1.15,
Domain.SOFTWARE_DEV : 1.15,
Domain.TECHNICAL_DOC : 1.10,
Domain.ENGINEERING : 1.10,
Domain.SCIENCE : 1.20,
Domain.BUSINESS : 1.05,
Domain.LEGAL : 1.25,
Domain.MEDICAL : 1.20,
Domain.JOURNALISM : 1.00,
Domain.MARKETING : 0.95,
Domain.SOCIAL_MEDIA : 0.80,
Domain.BLOG_PERSONAL : 0.90,
Domain.TUTORIAL : 1.00,
}
sensitivity = domain_sensitivity.get(domain, 1.0)
text_lower = text.lower()
for model, fingerprints in self.MODEL_FINGERPRINTS.items():
match_count = 0
total_checks = 0
# Check phrase matches
if ("phrases" in fingerprints):
for phrase in fingerprints["phrases"]:
if (phrase in text_lower):
match_count += 3
total_checks += 1
# Check sentence starters
if ("sentence_starters" in fingerprints):
sentences = re.split(r'[.!?]+', text)
for sentence in sentences:
sentence = sentence.strip().lower()
for starter in fingerprints["sentence_starters"]:
if (sentence.startswith(starter)):
match_count += 2
break
total_checks += len(sentences)
# Check structural patterns
if ("structural_patterns" in fingerprints):
for pattern in fingerprints["structural_patterns"]:
if (pattern in text_lower):
match_count += 2
total_checks += 1
# Calculate normalized score
if (total_checks > 0):
base_score = min(1.0, match_count / (total_checks * 0.5))
# Apply domain calibration
scores[model] = min(1.0, base_score * sensitivity)
return scores
def _analyze_statistical_patterns(self, text: str, domain: Domain) -> Dict[AIModel, float]:
"""
Analyze statistical patterns to identify model with domain awareness
"""
scores = {model: 0.3 for model in AIModel if model not in [AIModel.HUMAN, AIModel.UNKNOWN]}
# Calculate text statistics
sentences = re.split(r'[.!?]+', text)
sentences = [s.strip() for s in sentences if s.strip()]
words = text.split()
if not sentences or not words:
return scores
# Basic statistics
avg_sentence_length = len(words) / len(sentences)
word_count = len(words)
sentence_count = len(sentences)
# Punctuation frequencies
em_dash_freq = text.count('—') / word_count if word_count else 0
semicolon_freq = text.count(';') / word_count if word_count else 0
colon_freq = text.count(':') / word_count if word_count else 0
comma_freq = text.count(',') / word_count if word_count else 0
question_freq = text.count('?') / sentence_count if sentence_count else 0
exclamation_freq = text.count('!') / sentence_count if sentence_count else 0
# DOMAIN-AWARE: Adjust expectations based on domains
domain_adjustments = {Domain.GENERAL : 1.00,
Domain.ACADEMIC : 1.10,
Domain.CREATIVE : 0.95,
Domain.AI_ML : 1.05,
Domain.SOFTWARE_DEV : 1.05,
Domain.TECHNICAL_DOC : 1.05,
Domain.ENGINEERING : 1.05,
Domain.SCIENCE : 1.08,
Domain.BUSINESS : 1.00,
Domain.LEGAL : 1.12,
Domain.MEDICAL : 1.08,
Domain.JOURNALISM : 0.95,
Domain.MARKETING : 0.92,
Domain.SOCIAL_MEDIA : 0.85,
Domain.BLOG_PERSONAL : 0.95,
Domain.TUTORIAL : 1.00,
}
domain_factor = domain_adjustments.get(domain, 1.0)
# Compare against model fingerprints
for model, fingerprints in self.MODEL_FINGERPRINTS.items():
if ("style_markers" not in fingerprints) or ("punctuation_patterns" not in fingerprints):
continue
style = fingerprints["style_markers"]
punct = fingerprints["punctuation_patterns"]
match_score = 0.3
# Check sentence length with domain adjustment
if ("avg_sentence_length" in style):
min_len, max_len = style["avg_sentence_length"]
adjusted_min = min_len * domain_factor
adjusted_max = max_len * domain_factor
if (adjusted_min <= avg_sentence_length <= adjusted_max):
match_score += 0.25
# Check punctuation patterns
punctuation_checks = [("em_dash_frequency", em_dash_freq),
("semicolon_frequency", semicolon_freq),
("colon_frequency", colon_freq),
("comma_frequency", comma_freq),
("question_frequency", question_freq),
("exclamation_frequency", exclamation_freq),
]
for pattern_name, observed_freq in punctuation_checks:
if (pattern_name in punct):
min_freq, max_freq = punct[pattern_name]
if (min_freq <= observed_freq <= max_freq):
match_score += 0.08
scores[model] = min(1.0, match_score)
return scores
def _analyze_metric_patterns(self, metric_results: Dict[str, MetricResult], domain: Domain) -> Dict[AIModel, float]:
"""
Use all 6 metrics with proper weights for attribution
"""
scores = {model: 0.0 for model in AIModel if model not in [AIModel.HUMAN, AIModel.UNKNOWN]}
if not metric_results:
return scores
# DOMAIN-AWARE: Adjust metric sensitivity based on domain
domain_metric_weights = {Domain.GENERAL : {"perplexity": 1.0, "structural": 1.0, "entropy": 1.0, "semantic_analysis": 1.0, "linguistic": 1.0, "multi_perturbation_stability": 1.0},
Domain.ACADEMIC : {"perplexity": 1.2, "structural": 1.0, "entropy": 0.9, "semantic_analysis": 1.1, "linguistic": 1.3, "multi_perturbation_stability": 0.8},
Domain.TECHNICAL_DOC : {"perplexity": 1.2, "structural": 1.1, "entropy": 0.9, "semantic_analysis": 1.2, "linguistic": 1.1, "multi_perturbation_stability": 0.8},
Domain.AI_ML : {"perplexity": 1.3, "structural": 1.0, "entropy": 0.9, "semantic_analysis": 1.2, "linguistic": 1.2, "multi_perturbation_stability": 0.8},
Domain.SOFTWARE_DEV : {"perplexity": 1.2, "structural": 1.1, "entropy": 0.9, "semantic_analysis": 1.1, "linguistic": 1.0, "multi_perturbation_stability": 0.9},
Domain.ENGINEERING : {"perplexity": 1.2, "structural": 1.1, "entropy": 0.9, "semantic_analysis": 1.1, "linguistic": 1.2, "multi_perturbation_stability": 0.8},
Domain.SCIENCE : {"perplexity": 1.2, "structural": 1.0, "entropy": 0.9, "semantic_analysis": 1.2, "linguistic": 1.3, "multi_perturbation_stability": 0.8},
Domain.BUSINESS : {"perplexity": 1.1, "structural": 1.0, "entropy": 1.0, "semantic_analysis": 1.2, "linguistic": 1.1, "multi_perturbation_stability": 0.9},
Domain.LEGAL : {"perplexity": 1.2, "structural": 1.1, "entropy": 0.9, "semantic_analysis": 1.3, "linguistic": 1.3, "multi_perturbation_stability": 0.8},
Domain.MEDICAL : {"perplexity": 1.2, "structural": 1.0, "entropy": 0.9, "semantic_analysis": 1.2, "linguistic": 1.2, "multi_perturbation_stability": 0.8},
Domain.JOURNALISM : {"perplexity": 1.1, "structural": 1.0, "entropy": 1.0, "semantic_analysis": 1.1, "linguistic": 1.1, "multi_perturbation_stability": 0.9},
Domain.CREATIVE : {"perplexity": 0.9, "structural": 0.9, "entropy": 1.2, "semantic_analysis": 1.0, "linguistic": 1.3, "multi_perturbation_stability": 0.9},
Domain.MARKETING : {"perplexity": 1.0, "structural": 1.0, "entropy": 1.1, "semantic_analysis": 1.1, "linguistic": 1.2, "multi_perturbation_stability": 0.8},
Domain.SOCIAL_MEDIA : {"perplexity": 1.0, "structural": 0.8, "entropy": 1.3, "semantic_analysis": 0.9, "linguistic": 0.9, "multi_perturbation_stability": 0.9},
Domain.BLOG_PERSONAL : {"perplexity": 1.0, "structural": 0.9, "entropy": 1.2, "semantic_analysis": 1.0, "linguistic": 1.1, "multi_perturbation_stability": 0.8},
Domain.TUTORIAL : {"perplexity": 1.1, "structural": 1.0, "entropy": 1.0, "semantic_analysis": 1.1, "linguistic": 1.1, "multi_perturbation_stability": 0.9},
}
domain_weights = domain_metric_weights.get(domain, domain_metric_weights[Domain.GENERAL])
# PERPLEXITY ANALYSIS (25% weight)
if ("perplexity" in metric_results):
perplexity_result = metric_results["perplexity"]
overall_perplexity = perplexity_result.details.get("overall_perplexity", 50)
domain_weight = domain_weights.get("perplexity", 1.0)
# GPT models typically have lower perplexity
if (overall_perplexity < 25):
scores[AIModel.GPT_4] += 0.6 * self.METRIC_WEIGHTS["perplexity"] * domain_weight
scores[AIModel.GPT_4_TURBO] += 0.5 * self.METRIC_WEIGHTS["perplexity"] * domain_weight
elif (overall_perplexity < 35):
scores[AIModel.GPT_3_5] += 0.4 * self.METRIC_WEIGHTS["perplexity"] * domain_weight
scores[AIModel.GEMINI_PRO] += 0.3 * self.METRIC_WEIGHTS["perplexity"] * domain_weight
# STRUCTURAL ANALYSIS (15% weight)
if ("structural" in metric_results):
structural_result = metric_results["structural"]
burstiness = structural_result.details.get("burstiness_score", 0.5)
uniformity = structural_result.details.get("length_uniformity", 0.5)
domain_weight = domain_weights.get("structural", 1.0)
# Claude models show more structural consistency
if (uniformity > 0.7):
scores[AIModel.CLAUDE_3_OPUS] += 0.5 * self.METRIC_WEIGHTS["structural"] * domain_weight
scores[AIModel.CLAUDE_3_SONNET] += 0.4 * self.METRIC_WEIGHTS["structural"] * domain_weight
# SEMANTIC ANALYSIS (15% weight)
if ("semantic_analysis" in metric_results):
semantic_result = metric_results["semantic_analysis"]
coherence = semantic_result.details.get("coherence_score", 0.5)
consistency = semantic_result.details.get("consistency_score", 0.5)
domain_weight = domain_weights.get("semantic_analysis", 1.0)
# GPT-4 shows exceptional semantic coherence
if (coherence > 0.8):
scores[AIModel.GPT_4] += 0.7 * self.METRIC_WEIGHTS["semantic_analysis"] * domain_weight
scores[AIModel.GPT_4_TURBO] += 0.6 * self.METRIC_WEIGHTS["semantic_analysis"] * domain_weight
# ENTROPY ANALYSIS (20% weight)
if ("entropy" in metric_results):
entropy_result = metric_results["entropy"]
token_diversity = entropy_result.details.get("token_diversity", 0.5)
sequence_unpredictability = entropy_result.details.get("sequence_unpredictability", 0.5)
domain_weight = domain_weights.get("entropy", 1.0)
# Higher entropy diversity suggests more sophisticated models
if (token_diversity > 0.7):
scores[AIModel.CLAUDE_3_OPUS] += 0.6 * self.METRIC_WEIGHTS["entropy"] * domain_weight
scores[AIModel.GPT_4] += 0.5 * self.METRIC_WEIGHTS["entropy"] * domain_weight
# LINGUISTIC ANALYSIS (15% weight)
if ("linguistic" in metric_results):
linguistic_result = metric_results["linguistic"]
pos_diversity = linguistic_result.details.get("pos_diversity", 0.5)
syntactic_complexity = linguistic_result.details.get("syntactic_complexity", 2.5)
domain_weight = domain_weights.get("linguistic", 1.0)
# Complex linguistic patterns suggest advanced models
if (syntactic_complexity > 3.0):
scores[AIModel.CLAUDE_3_OPUS] += 0.5 * self.METRIC_WEIGHTS["linguistic"] * domain_weight
scores[AIModel.GPT_4] += 0.4 * self.METRIC_WEIGHTS["linguistic"] * domain_weight
# MULTI-PERTURBATION STABILITY ANALYSIS (10% weight)
if ("multi_perturbation_stability" in metric_results):
multi_perturbation_stability_result = metric_results["multi_perturbation_stability"]
stability = multi_perturbation_stability_result.details.get("stability_score", 0.5)
curvature = multi_perturbation_stability_result.details.get("curvature_score", 0.5)
# Specific stability patterns for different model families
if (0.4 <= stability <= 0.6):
scores[AIModel.MIXTRAL] += 0.4 * self.METRIC_WEIGHTS["multi_perturbation_stability"]
scores[AIModel.LLAMA_3] += 0.3 * self.METRIC_WEIGHTS["multi_perturbation_stability"]
# Normalize scores
for model in scores:
scores[model] = min(1.0, scores[model])
return scores
def _combine_attribution_scores(self, fingerprint_scores: Dict[AIModel, float], statistical_scores: Dict[AIModel, float],
metric_scores: Dict[AIModel, float], domain: Domain) -> Tuple[Dict[str, float], Dict[str, float]]:
"""
ENSEMBLE COMBINATION using document-specified weights and domain awareness
"""
# DOMAIN-AWARE weighting for ALL 16 DOMAINS
domain_weights = {Domain.GENERAL : {"fingerprint": 0.35, "statistical": 0.30, "metric": 0.35},
Domain.ACADEMIC : {"fingerprint": 0.30, "statistical": 0.35, "metric": 0.35},
Domain.TECHNICAL_DOC : {"fingerprint": 0.25, "statistical": 0.40, "metric": 0.35},
Domain.AI_ML : {"fingerprint": 0.28, "statistical": 0.37, "metric": 0.35},
Domain.SOFTWARE_DEV : {"fingerprint": 0.27, "statistical": 0.38, "metric": 0.35},
Domain.ENGINEERING : {"fingerprint": 0.28, "statistical": 0.37, "metric": 0.35},
Domain.SCIENCE : {"fingerprint": 0.30, "statistical": 0.35, "metric": 0.35},
Domain.BUSINESS : {"fingerprint": 0.33, "statistical": 0.35, "metric": 0.32},
Domain.LEGAL : {"fingerprint": 0.28, "statistical": 0.40, "metric": 0.32},
Domain.MEDICAL : {"fingerprint": 0.30, "statistical": 0.38, "metric": 0.32},
Domain.JOURNALISM : {"fingerprint": 0.35, "statistical": 0.33, "metric": 0.32},
Domain.CREATIVE : {"fingerprint": 0.40, "statistical": 0.30, "metric": 0.30},
Domain.MARKETING : {"fingerprint": 0.38, "statistical": 0.32, "metric": 0.30},
Domain.SOCIAL_MEDIA : {"fingerprint": 0.45, "statistical": 0.35, "metric": 0.20},
Domain.BLOG_PERSONAL : {"fingerprint": 0.42, "statistical": 0.32, "metric": 0.26},
Domain.TUTORIAL : {"fingerprint": 0.33, "statistical": 0.35, "metric": 0.32},
}
weights = domain_weights.get(domain, domain_weights[Domain.GENERAL])
combined = dict()
metric_contributions = dict()
all_models = set(fingerprint_scores.keys()) | set(statistical_scores.keys()) | set(metric_scores.keys())
for model in all_models:
score = (fingerprint_scores.get(model, 0.0) * weights["fingerprint"] +
statistical_scores.get(model, 0.0) * weights["statistical"] +
metric_scores.get(model, 0.0) * weights["metric"]
)
combined[model.value] = score
# Normalize scores to sum to 1.0 for proper probability distribution
total_score = sum(combined.values())
if (total_score > 0):
combined = {model: score / total_score for model, score in combined.items()}
# Calculate metric contributions for explainability
if metric_scores:
total_metric_impact = sum(metric_scores.values())
if (total_metric_impact > 0):
for model, score in metric_scores.items():
metric_contributions[model.value] = score / total_metric_impact
return combined, metric_contributions
def _make_domain_aware_prediction(self, combined_scores: Dict[str, float], domain: Domain, domain_preferences: List[AIModel]) -> Tuple[AIModel, float]:
"""
Domain aware prediction that considers domain-specific model preferences
"""
if not combined_scores:
return AIModel.UNKNOWN, 0.0
# Find the model with the highest probability
sorted_models = sorted(combined_scores.items(), key=lambda x: x[1], reverse=True)
if not sorted_models:
return AIModel.UNKNOWN, 0.0
best_model_name, best_score = sorted_models[0]
# Thresholding to show model only if confidence is sufficient
if (best_score < 0.01):
return AIModel.UNKNOWN, best_score
try:
best_model = AIModel(best_model_name)
except ValueError:
best_model = AIModel.UNKNOWN
# Calculate confidence - be more generous
if (len(sorted_models) > 1):
second_score = sorted_models[1][1]
margin = best_score - second_score
# More generous confidence calculation
confidence = min(1.0, best_score * 0.8 + margin * 1.5)
else:
confidence = best_score * 0.9
# Always return the actual best model, never downgrade to UNKNOWN
return best_model, max(0.05, confidence)
def _generate_detailed_reasoning(self, predicted_model: AIModel, confidence: float, domain: Domain, metric_contributions: Dict[str, float],
combined_scores: Dict[str, float]) -> List[str]:
"""
Generate Explainable reasoning - ENHANCED version
"""
reasoning = []
reasoning.append("**AI Model Attribution Analysis**")
reasoning.append("")
# Show prediction with confidence
if (predicted_model == AIModel.UNKNOWN):
reasoning.append("**Most Likely**: Unable to determine with high confidence")
else:
model_name = predicted_model.value.replace("-", " ").replace("_", " ").title()
reasoning.append(f"**Predicted Model**: {model_name}")
reasoning.append(f"**Confidence**: {confidence*100:.1f}%")
reasoning.append(f"**Domain**: {domain.value.replace('_', ' ').title()}")
reasoning.append("")
# Show model probability distribution
reasoning.append("**Model Probability Distribution:**")
reasoning.append("")
if combined_scores:
sorted_models = sorted(combined_scores.items(), key = lambda x: x[1], reverse = True)
for i, (model_name, score) in enumerate(sorted_models[:6]):
# Skip very low probabilities
if (score < 0.01):
continue
display_name = model_name.replace("-", " ").replace("_", " ").title()
percentage = score * 100
# Use proper markdown formatting
reasoning.append(f"• **{display_name}**: {percentage:.1f}%")
reasoning.append("")
# Add analysis insights
reasoning.append("**Analysis Notes:**")
if (confidence < 0.3):
reasoning.append("• Low confidence attribution - text patterns are ambiguous")
reasoning.append("• May be human-written or from multiple AI sources")
else:
reasoning.append(f"• Calibrated for {domain.value.replace('_', ' ')} domain")
# Domain-specific insights
domain_insights = {Domain.ACADEMIC : "Academic writing patterns analyzed",
Domain.TECHNICAL_DOC : "Technical coherence and structure weighted",
Domain.CREATIVE : "Stylistic and linguistic diversity emphasized",
Domain.SOCIAL_MEDIA : "Casual language and engagement patterns considered",
Domain.AI_ML : "Technical terminology and analytical patterns emphasized",
Domain.SOFTWARE_DEV : "Code-like structures and technical precision weighted",
Domain.ENGINEERING : "Technical specifications and formal language analyzed",
Domain.SCIENCE : "Scientific terminology and methodological patterns considered",
Domain.BUSINESS : "Professional communication and strategic language weighted",
Domain.LEGAL : "Formal language and legal terminology emphasized",
Domain.MEDICAL : "Medical terminology and clinical language analyzed",
Domain.JOURNALISM : "News reporting style and factual presentation weighted",
Domain.MARKETING : "Persuasive language and engagement patterns considered",
Domain.BLOG_PERSONAL : "Personal voice and conversational style analyzed",
Domain.TUTORIAL : "Instructional clarity and step-by-step structure weighted",
}
insight = domain_insights.get(domain, "Multiple attribution factors analyzed")
reasoning.append(f"• {insight}")
return reasoning
def _get_top_fingerprints(self, fingerprint_scores: Dict[AIModel, float]) -> Dict[str, int]:
"""
Get top fingerprint matches for display
"""
top_matches = dict()
sorted_models = sorted(fingerprint_scores.items(), key = lambda x: x[1], reverse = True)[:5]
for model, score in sorted_models:
# Only show meaningful matches
if (score > 0.1):
top_matches[model.value] = int(score * 100)
return top_matches
def _create_unknown_result(self, domain: Domain) -> AttributionResult:
"""
Create result for unknown attribution with domain context
"""
return AttributionResult(predicted_model = AIModel.UNKNOWN,
confidence = 0.0,
model_probabilities = {},
reasoning = [f"Model attribution inconclusive for {domain.value} content. Text may be human-written or from unidentifiable model"],
fingerprint_matches = {},
domain_used = domain,
metric_contributions = {},
)
# Export
__all__ = ["AIModel",
"ModelAttributor",
"AttributionResult",
] |