Spaces:
Sleeping
Sleeping
File size: 5,342 Bytes
306e475 357db8c 306e475 357db8c 306e475 357db8c 306e475 357db8c 306e475 357db8c 306e475 357db8c 306e475 357db8c 306e475 357db8c 306e475 357db8c 306e475 357db8c 306e475 357db8c 306e475 357db8c 306e475 357db8c | 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 | import torch
from transformers import pipeline
import logging
# -- LOGGER ---
logger = logging.getLogger(__name__)
class LegalNLPEngine:
def __init__(self):
self.model_name = "recognai/zeroshot_selectra_medium"
self.device = 0 if torch.cuda.is_available() else -1
print(f"Loading NLP Model: {self.model_name} on device {self.device}...")
try:
self.classifier = pipeline(
"zero-shot-classification", model=self.model_name, device=self.device
)
except Exception as e:
logger.error(f"Error loading model: {e}")
self.classifier = None
def analyze_clause(self, text: str):
if not text or len(text) < 15:
return None
text_lower = text.lower()
# --- LEVEL 1: RISK HEURISTIC ---
risky_keywords = [
# --- Bloque: Renuncias y Legal ---
"modificaci贸n unilateral",
"modificar unilateralmente",
"exenci贸n de responsabilidad",
"no se hace responsable",
"renuncia a derechos",
"renuncia de forma expresa",
"renuncia expresa",
"irrevocable",
"renuncia al fuero",
"renuncia a cualquier otro fuero",
"juzgados que designe la empresa",
"juzgados que libremente designe", #
# --- Bloque: Condiciones Laborales ---
"sin preaviso",
"sin necesidad de causa",
"sin necesidad de alegar causa",
"sin derecho a compensaci贸n",
"sin compensaci贸n econ贸mica",
"no genera derecho",
"absorbe cualquier concepto",
"cualesquiera otras tareas",
"no guarden relaci贸n directa",
# --- Bloque: Movilidad y Funciones ---
"movilidad geogr谩fica",
"traslado a cualquier",
"podr谩 trasladar",
"cambio de centro",
"funciones de distinta categor铆a",
"polivalencia funcional",
# --- Bloque: Tiempo y Vacaciones ---
"jornada de hasta",
"horas extraordinarias obligatorias",
"realizaci贸n ilimitada",
"disponibilidad total",
"cancelar las vacaciones",
"modificar las vacaciones",
"fraccionar las vacaciones",
"fijada exclusivamente por la empresa",
# --- Bloque: Pagos ---
"cuando su tesorer铆a",
"retrasarlo hasta",
"pago diferido",
"sin que ello genere intereses",
# --- Bloque: Privacidad y Sanciones ---
"despido disciplinario inmediato",
"comentarios privados",
"uso ilimitado de su imagen",
"cesi贸n de imagen",
"datos a terceros"
]
for keyword in risky_keywords:
if keyword in text_lower:
return {
"text_snippet": text[:150] + "...",
"label": "POTENTIAL_RISK",
"confidence": 0.98,
"is_risky": True,
}
# --- LEVEL 2: FILTER "ADMINISTRATIVE NOISE" ---
safe_keywords = [
"en madrid a",
"reunidos",
"con domicilio en",
"con dni",
"mayor de edad",
"intervienen",
"exponen",
"cl谩usulas:",
"firmado en",
"fdo.",
"el trabajador:",
"la empresa:",
]
if any(sk in text_lower for sk in safe_keywords):
return {
"text_snippet": text[:150] + "...",
"label": "ACCEPTABLE",
"confidence": 0.90,
"is_risky": False,
}
# --- LEVEL 3: ARTIFICIAL INTELLIGENCE (Zero-Shot) ---
if self.classifier:
try:
candidate_labels = [
"cl谩usula abusiva",
"explotaci贸n laboral",
"renuncia de derechos",
"condici贸n laboral est谩ndar",
"informaci贸n administrativa",
]
result = self.classifier(text, candidate_labels)
top_label = result["labels"][0]
score = result["scores"][0]
risky_labels = [
"cl谩usula abusiva",
"explotaci贸n laboral",
"renuncia de derechos",
]
is_risky_ai = top_label in risky_labels and score > 0.40
return {
"text_snippet": text[:150] + "...",
"label": "AI_DETECTED_RISK" if is_risky_ai else "ACCEPTABLE",
"confidence": round(score, 2),
"is_risky": is_risky_ai,
}
except Exception as e:
logger.error(f"AI Inference error: {e}")
# Fallback
return {
"text_snippet": text[:100] + "...",
"label": "NEUTRAL",
"confidence": 0.0,
"is_risky": False,
}
# Singleton instance
nlp_engine = LegalNLPEngine()
|