| """ |
| Módulo de control de calidad y normalización de eventos y entidades. |
| |
| Incluye: |
| 1. Herramientas para limpiar texto y validar eventos (5W1H). |
| 2. Lógica de deduplicación de entidades. |
| 3. Auditoría de calidad para Entity Linking. |
| """ |
|
|
| import logging |
| import unicodedata |
| import string |
| from urllib.parse import unquote |
| from typing import List, Dict, Any |
|
|
| logger = logging.getLogger(__name__) |
|
|
| |
| |
| |
|
|
| def clean_punctuation(text: str) -> str: |
| """ |
| Reemplaza puntuación por ESPACIOS. |
| Crucial para casos como 'dijo,que' -> 'dijo que'. |
| """ |
| if not isinstance(text, str): return "" |
| |
| translator = str.maketrans(string.punctuation, ' ' * len(string.punctuation)) |
| return text.translate(translator) |
|
|
|
|
| def normalize_text(text: str) -> str: |
| """ |
| Normalización base para comparación de texto (sin tildes, minúsculas). |
| """ |
| if not isinstance(text, str): return "" |
| |
| normalized = unicodedata.normalize('NFKD', text) |
| ascii_text = normalized.encode('ascii', 'ignore').decode('ascii') |
| return ascii_text.lower().strip() |
|
|
|
|
| def generate_slug(text: str) -> str: |
| """ |
| Específico para crear IDs de Neo4j (local:pedro_sanchez). |
| Usa la normalización base pero asegura guiones bajos. |
| """ |
| |
| clean = normalize_text(text) |
| |
| clean = clean.translate(str.maketrans('', '', string.punctuation)) |
| |
| return clean.replace(' ', '_') |
|
|
|
|
| |
| |
| |
|
|
| |
| _STOP_VERBS_RAW = { |
| 'ser', 'estar', 'parecer', 'haber', 'tener', 'hacer', 'ir', |
| 'poder', 'deber', 'querer', 'decir', 'señalar', 'indicar', |
| 'afirmar', 'explicar', 'asegurar', 'manifestar', 'comentar', |
| 'es', 'son', 'era', 'fue', 'fueron', 'sido', 'siendo', |
| 'está', 'están', 'estaba', 'estuvo', 'estuvieron', |
| 'ha', 'han', 'había', 'hubo', 'habido', |
| 'dijo', 'dijeron', 'dice', 'dicen', |
| 'señaló', 'señalaron', 'señala', |
| 'afirmó', 'afirmaron', 'afirma', |
| 'explicó', 'explicaron', 'explica', |
| 'aseguró', 'aseguraron', 'asegura', |
| 'manifestó', 'manifestaron', 'manifiesta', |
| 'comentó', 'comentaron', 'comenta' |
| } |
| |
| STOP_VERBS = {normalize_text(v) for v in _STOP_VERBS_RAW} |
|
|
| |
| PARTICLES = {'se', 'no', 'le', 'les', 'lo', 'la', 'los', 'las', 'me', 'te', 'nos', 'os'} |
|
|
| |
| INVALID_WHO_PATTERNS = { |
| |
| 'que', 'quien', 'quienes', 'cual', 'cuales', 'cuyo', 'cuya', |
| |
| 'eso', 'esto', 'aquello', 'ello', |
| |
| 'algo', 'alguien', 'nadie', 'nada', 'todo', 'todos', 'ambos', |
| |
| 'yo', 'tu', 'nosotros', 'vosotros', 'usted', 'ustedes' |
| } |
|
|
| |
| UNICODE_PUNCTUATION = '—–«»""''…•·¿¡' |
|
|
|
|
| |
| |
| |
|
|
| def get_main_verb(what_span: str) -> str: |
| """Extrae el verbo principal limpiando puntuación y saltando partículas.""" |
| clean_span = clean_punctuation(what_span) |
| words = clean_span.split() |
| for word in words: |
| word_normalized = normalize_text(word) |
| if word_normalized not in PARTICLES: |
| return word_normalized |
| return words[0] if words else '' |
|
|
|
|
| def is_valid_who(who: Dict[str, Any]) -> bool: |
| """Filtro para el campo WHO.""" |
| span = who.get('span', '').strip() |
| has_uri = who.get('uri') is not None |
| ner_type = who.get('type') |
|
|
| if len(span) < 2 or len(span) > 60: return False |
| if span[0] in string.punctuation or span[0] in UNICODE_PUNCTUATION: return False |
| if span[-1] in string.punctuation or span[-1] in UNICODE_PUNCTUATION: return False |
| if ',' in span: return False |
|
|
| clean_span = clean_punctuation(span) |
| words = clean_span.split() |
| if len(words) > 5: return False |
|
|
| if words: |
| first_word = normalize_text(words[0]) |
| if first_word in INVALID_WHO_PATTERNS: return False |
|
|
| if not has_uri and ner_type not in ['PER', 'ORG', 'LOC']: return False |
|
|
| return True |
|
|
|
|
| def is_valid_event(event: Dict[str, Any]) -> bool: |
| """Aplica filtro de calidad al evento completo.""" |
| has_valid_who = False |
| for who in event.get('who', []): |
| if not is_valid_who(who): continue |
| if who.get('uri') or who.get('type') in ['PER', 'ORG', 'LOC']: |
| has_valid_who = True |
| break |
|
|
| if not has_valid_who: return False |
|
|
| what_list = event.get('what', []) |
| if not what_list or what_list[0].get('start', -1) == -1: return False |
|
|
| what_span = what_list[0].get('span', '') |
| main_verb = get_main_verb(what_span) |
|
|
| if main_verb in STOP_VERBS: return False |
| if len(what_span) < 3 or len(what_span) > 200: return False |
|
|
| return True |
|
|
|
|
| def deduplicate_entities(entities_list: List[Dict[str, Any]]) -> Dict[str, Dict[str, Any]]: |
| """Deduplicación de entidades basada en frecuencia y completitud.""" |
| unique_entities = {} |
| name_frequency = {} |
|
|
| for ent in entities_list: |
| uri = ent.get('uri') |
| raw_name = ent.get('span', '').strip() |
|
|
| if not uri: |
| uri = f"local:{generate_slug(raw_name)}" |
|
|
| if uri not in unique_entities: |
| unique_entities[uri] = { |
| 'name': raw_name, |
| 'type': ent.get('type'), |
| 'uri': uri, |
| 'mention_count': 0 |
| } |
| name_frequency[uri] = {} |
|
|
| if raw_name not in name_frequency[uri]: |
| name_frequency[uri][raw_name] = 0 |
| name_frequency[uri][raw_name] += 1 |
| unique_entities[uri]['mention_count'] += 1 |
|
|
| if not unique_entities[uri]['type'] and ent.get('type'): |
| unique_entities[uri]['type'] = ent.get('type') |
|
|
| |
| for uri in unique_entities: |
| if not name_frequency[uri]: continue |
| total_mentions = unique_entities[uri]['mention_count'] |
| most_frequent = max(name_frequency[uri], key=name_frequency[uri].get) |
| best_name = most_frequent |
| base_normalized = normalize_text(most_frequent) |
|
|
| for candidate, count in name_frequency[uri].items(): |
| if len(candidate) <= len(best_name): continue |
| if base_normalized not in normalize_text(candidate): continue |
| if count >= total_mentions * 0.10: |
| best_name = candidate |
|
|
| unique_entities[uri]['name'] = best_name |
|
|
| return unique_entities |
|
|
|
|
| |
| |
| |
|
|
| KNOWN_DISAMBIGUATION_ERRORS = { |
| "Homo_erectus_pekinensis": "Falso positivo crítico: Capital (Pekín) -> Especie homínida", |
| "Bandera_de_Europa": "Error metonímico: Organización (UE) -> Símbolo (Bandera)", |
| "Bandera_de_la_República_Popular_China": "Error metonímico: País (China) -> Símbolo (Bandera)", |
| "Agencia_Espacial_Europea": "Ambigüedad contextual: Pronombre ('esa') -> Organización (ESA)", |
| "Mariano_Rajoy": "Asociación obsoleta: Cargo (Presidente) -> Entidad histórica", |
| "Parlamento_Europeo": "Desviación semántica: Organización relacionada -> Institución específica", |
| "Hoja": "Ruido: Sustantivo común -> Entidad botánica", |
| "Cerdo": "Ruido: Sustantivo común -> Entidad biológica", |
| "Masa": "Ruido: Magnitud física -> Entidad", |
| "Guerra_comercial": "Generalización: Concepto abstracto -> Evento histórico" |
| } |
|
|
| SUSPICIOUS_TYPES = { |
| "DBpedia:Eukaryote", "DBpedia:Mammal", "DBpedia:Animal", |
| "DBpedia:Fungus", "DBpedia:Species", "DBpedia:Biomolecule" |
| } |
|
|
| SAFE_TYPES = { |
| "DBpedia:Person", "DBpedia:Place", "DBpedia:Organisation", |
| "DBpedia:Country", "DBpedia:City", "DBpedia:Company" |
| } |
|
|
| def audit_linking_quality(text: str, linker: Any, thresholds: List[float]) -> None: |
| for confidence in thresholds: |
| print(f"--- Evaluando con umbral: {confidence} ---") |
| try: |
| annotations = linker.annotate_text(text, confidence=confidence) |
| count = len(annotations) |
| print(f" > Entidades encontradas: {count}") |
|
|
| if count > 0: |
| _analyze_anomalies(annotations) |
| else: |
| print(" > Sin resultados.") |
| |
| except Exception as e: |
| import traceback |
| traceback.print_exc() |
| print(f" > Error en ejecución: {str(e)}") |
| print("") |
|
|
| def _analyze_anomalies(annotations: List[Dict[str, Any]]) -> None: |
| detected_problems = [] |
| |
| for entity in annotations: |
| raw_uri = entity.get('uri', '') |
| text = entity.get('text', '') |
| decoded_uri = unquote(raw_uri) |
| uri_suffix = decoded_uri.split('/')[-1] |
| |
| if uri_suffix in KNOWN_DISAMBIGUATION_ERRORS: |
| description = KNOWN_DISAMBIGUATION_ERRORS[uri_suffix] |
| detected_problems.append(f"[URI PROHIBIDA] '{text}' -> {uri_suffix} ({description})") |
| continue |
|
|
| raw_types = entity.get('types', []) |
| if isinstance(raw_types, list): |
| entity_types = set(raw_types) |
| elif isinstance(raw_types, str): |
| entity_types = set(raw_types.split(',')) if raw_types else set() |
| else: |
| entity_types = set() |
|
|
| suspicious_matches = entity_types.intersection(SUSPICIOUS_TYPES) |
| has_safe_type = not entity_types.isdisjoint(SAFE_TYPES) |
| |
| if suspicious_matches and not has_safe_type: |
| detected_problems.append(f"[TIPO SOSPECHOSO] '{text}' -> {uri_suffix} ({', '.join(suspicious_matches)})") |
|
|
| if detected_problems: |
| print(" [!] ANOMALÍAS DETECTADAS:") |
| for problem in sorted(list(set(detected_problems))): |
| print(f" * {problem}") |
| else: |
| print(" > No se detectaron anomalías obvias en esta muestra.") |