|
|
|
|
|
|
|
|
|
import time
|
|
from typing import Dict, List, Optional, Any
|
|
|
|
|
|
try:
|
|
from aee_core_classes_era import Proposition, EpistemicData
|
|
from aee_extractor_era import process_with_spacy, extract_propositions_era, NLP_MODEL
|
|
from aee_linker_era import find_and_link_evidence_era
|
|
from aee_updater_era import run_updates_era
|
|
from aee_explainer_era import generate_explanation_era
|
|
from aee_bias_detector import run_bias_detection_v3
|
|
from aee_validator import check_plausibility_v_era
|
|
from aee_utils import get_proposition_by_id
|
|
except ImportError as e:
|
|
print(f"Fatal Error: Could not import necessary modules. Check file paths and dependencies in AEE/Era folder.")
|
|
print(f"Import Error: {e}")
|
|
exit()
|
|
|
|
|
|
def report_kb_era(kb: Dict[str, Proposition]):
|
|
|
|
print("\n" + "="*70); print(" AEE Era Version - Knowledge Base Report (Final Status)"); print("="*70)
|
|
if not kb: print("Knowledge Base is empty."); print("="*70); return
|
|
print(f"Total propositions in KB: {len(kb)}"); print("-"*70)
|
|
propositions_by_source: Dict[str, List[Proposition]] = {};
|
|
def sort_key(prop): conf = prop.epistemic_data.computed_confidence; return conf if conf is not None else -1.0
|
|
sorted_props = sorted(list(kb.values()), key=sort_key, reverse=True)
|
|
for prop in sorted_props: source = prop.epistemic_data.source_id; propositions_by_source.setdefault(source, []).append(prop)
|
|
for source_id in sorted(propositions_by_source.keys()):
|
|
props = propositions_by_source[source_id]; source_reliability = getattr(props[0].epistemic_data, 'reliability_score', None)
|
|
reliability_str = f"{source_reliability:.2f}" if source_reliability is not None else "N/A"
|
|
print(f"\n--- Source: {source_id} (Calculated Reliability: {reliability_str}) ---")
|
|
for prop in props:
|
|
neg_str = "[NEGATED] " if prop.is_negated else ""
|
|
supports_str = ', '.join([pid[:8] for pid in prop.epistemic_data.supports]) if prop.epistemic_data.supports else "None"
|
|
contradicts_str = ', '.join([pid[:8] for pid in prop.epistemic_data.contradicts]) if prop.epistemic_data.contradicts else "None"
|
|
bias_str = ', '.join(prop.epistemic_data.bias_flags) if prop.epistemic_data.bias_flags else "None"
|
|
plausibility_score = prop.epistemic_data.plausibility_score; plausibility_str = f"{plausibility_score:.2f}" if plausibility_score is not None else "N/A"
|
|
validation_notes_str = ', '.join(prop.epistemic_data.validation_notes) if prop.epistemic_data.validation_notes else "None"
|
|
conf_score = prop.epistemic_data.computed_confidence; conf_str = f"{conf_score:.3f}" if conf_score is not None else "N/A"
|
|
init_conf_score = prop.epistemic_data.initial_confidence; init_conf_str = f"{init_conf_score:.2f}" if init_conf_score is not None else "N/A"
|
|
print(f" Prop ID : {prop.prop_id}")
|
|
print(f" Struct: {neg_str}{prop.subject_lemma} - {prop.relation_lemma} - {prop.value_lemma}")
|
|
print(f" Conf. : {conf_str} (Initial: {init_conf_str})")
|
|
print(f" Links : Supports: [{supports_str}] | Contradicts: [{contradicts_str}]")
|
|
print(f" Biases: [{bias_str}]")
|
|
print(f" Plaus.: {plausibility_str} | Notes: [{validation_notes_str}]")
|
|
print("\n" + "="*70); print(" End of KB Report "); print("="*70)
|
|
|
|
|
|
|
|
def run_aee_era_pipeline(inputs: List[Dict[str, str]]) -> Dict[str, Proposition]:
|
|
"""
|
|
Verilen girdiler için AEE Era işlem hattını tam olarak çalıştırır
|
|
(Era Extract, Plausibility Check, Era Linker, Bias Detect, Era Update).
|
|
"""
|
|
if NLP_MODEL is None: print("FATAL ERROR: spaCy model not loaded."); return {}
|
|
|
|
print("\nStarting AEE Era Final Pipeline...")
|
|
knowledge_base: Dict[str, Proposition] = {}
|
|
start_time = time.time()
|
|
|
|
|
|
print("Phase 1: Extracting(Era), Validating Plausibility, and Linking(Era)...")
|
|
all_extracted_props_before_linking: List[Proposition] = []
|
|
for item in inputs:
|
|
source_id = item.get("source_id", f"unknown_source_{int(time.time())}"); text = item.get("text", "")
|
|
if not text: continue
|
|
doc = process_with_spacy(text)
|
|
if doc:
|
|
|
|
extracted_props = extract_propositions_era(doc, source_id)
|
|
for prop in extracted_props:
|
|
plausibility_score, validation_notes = check_plausibility_v_era(prop)
|
|
if hasattr(prop, 'epistemic_data') and prop.epistemic_data:
|
|
prop.epistemic_data.plausibility_score = plausibility_score
|
|
if validation_notes: prop.epistemic_data.validation_notes.extend(validation_notes)
|
|
all_extracted_props_before_linking.append(prop)
|
|
print(f" Phase 1a (Extraction(Era) & Validation) complete. Total extracted: {len(all_extracted_props_before_linking)}")
|
|
|
|
print(" Phase 1b (Linking(Era))...")
|
|
if find_and_link_evidence_era:
|
|
for new_prop in all_extracted_props_before_linking:
|
|
if new_prop.prop_id not in knowledge_base:
|
|
|
|
find_and_link_evidence_era(new_prop, knowledge_base)
|
|
knowledge_base[new_prop.prop_id] = new_prop
|
|
else: print("Skipping linking due to import error.")
|
|
print(f"Phase 1 (Extract(Era), Validate, Link(Era)) complete. KB size: {len(knowledge_base)}")
|
|
|
|
|
|
print("\nPhase 1.5: Running Bias Detection Heuristics...")
|
|
if run_bias_detection_v3 and knowledge_base: run_bias_detection_v3(knowledge_base)
|
|
else: print("Skipping Bias Detection due to import error or empty KB.")
|
|
print("Phase 1.5 complete.")
|
|
|
|
|
|
print("\nPhase 2: Running Era Updates (Reliability, Cycle Detect, Plausibility-aware Confidence)...")
|
|
if run_updates_era: updated_knowledge_base = run_updates_era(knowledge_base)
|
|
else: print("Skipping Updates due to import error."); updated_knowledge_base = knowledge_base
|
|
print("Phase 2 complete.")
|
|
|
|
end_time = time.time(); print(f"\nPipeline finished in {end_time - start_time:.2f} seconds.")
|
|
return updated_knowledge_base
|
|
|
|
|
|
if __name__ == "__main__":
|
|
|
|
sample_inputs_era_final = [
|
|
{
|
|
"source_id": "fact_sheet_1", "source_type": "fact",
|
|
"text": "Water is H2O. The sun is hot. Ice is cold."
|
|
},
|
|
{
|
|
"source_id": "opinion_blog_A", "source_type": "blog",
|
|
"text": "Maybe the new policy is good. It could improve things. Perhaps."
|
|
},
|
|
{
|
|
"source_id": "opinion_blog_B", "source_type": "blog",
|
|
"text": "The new policy is definitely bad! It will undoubtedly harm the economy. It is not good."
|
|
},
|
|
{
|
|
"source_id": "report_X", "source_type": "report",
|
|
"text": "System Alpha is bigger than System Beta. System Beta is not small compared to Alpha."
|
|
},
|
|
{
|
|
"source_id": "another_report", "source_type": "report",
|
|
"text": "System Alpha is large."
|
|
},
|
|
{
|
|
"source_id": "conspiracy_theory.blog", "source_type": "blog",
|
|
"text": "The moon landing was faked because the photos look wrong. The photos look wrong because the shadows are incorrect. The shadows are incorrect because the landing was faked."
|
|
|
|
|
|
"Also, the moon is made of cheese."
|
|
}
|
|
]
|
|
|
|
|
|
final_kb_era = run_aee_era_pipeline(sample_inputs_era_final)
|
|
|
|
|
|
report_kb_era(final_kb_era)
|
|
|
|
|
|
print("\n" + "#"*70)
|
|
print(" AEE Era Version - Generating Explanations")
|
|
print("#"*70)
|
|
if final_kb_era and generate_explanation_era:
|
|
ids_to_explain = list(final_kb_era.keys())
|
|
print(f"\nGenerating explanations for all {len(ids_to_explain)} propositions...\n")
|
|
for prop_id in ids_to_explain:
|
|
explanation = generate_explanation_era(prop_id, final_kb_era)
|
|
print(explanation)
|
|
print("-" * 40)
|
|
else: print("Knowledge Base is empty or Explainer not available.")
|
|
print("\n" + "#"*70); print(" Explanation generation step complete."); print("#"*70)
|
|
|
|
|
|
print("\n###########################################################################")
|
|
print("# AEE ERA VERSION - ALL PLANNED CORE FEATURE CODING COMPLETE!")
|
|
print("# All modules updated to Era versions where planned.")
|
|
print("# Project includes: Extraction(Era.2a), Validation(Era.1), Linking(Era.2b),")
|
|
print("# Bias Detection(v3), Updates(Era.1d), Explanation(Era.1e).")
|
|
print("#")
|
|
print("# FINAL STEP (FOR YOU): TESTING & EVALUATION!")
|
|
print("# - Run this script: python aee_era_main.py")
|
|
print("# - Examine the report and explanations thoroughly.")
|
|
print("# - Check if initial confidence reflects modality.")
|
|
print("# - Check if more links (support/contradiction) are found.")
|
|
print("# - Check bias flags, plausibility, final confidence.")
|
|
print("# - Try your own texts!")
|
|
print("# - Provide your final feedback and evaluation.")
|
|
print("###########################################################################") |