Update src/engine.py

src/engine.py

@@ -1,42 +1,148 @@
 import spacy
 import nltk
+
 from nltk.sentiment.vader import SentimentIntensityAnalyzer
-from spacy.cli import download
+
 
 class DeepPragmaEngine:
+    """
+    Motor local de clasificación pragmática.
+    Detecta:
+    - Hate Speech directo
+    - Hate Speech sarcástico
+    - Ironía no ofensiva
+    - Neutral
+    """
+
     def __init__(self):
-        # Configuración de NLTK
+
+        # Comprobación de recursos NLTK
         try:
-            nltk.data.find('sentiment/vader_lexicon.zip')
+            nltk.data.find("sentiment/vader_lexicon")
         except LookupError:
-            nltk.download('vader_lexicon')
-            
-        # Configuración de SpaCy con autodescarga
-        model_name = "en_core_web_sm"
+            raise RuntimeError(
+                "No se encontró vader_lexicon. "
+                "Instálalo durante el build con: "
+                "python -c \"import nltk; nltk.download('vader_lexicon')\""
+            )
+
+        # Carga del modelo spaCy
         try:
-            self.nlp = spacy.load(model_name)
+            self.nlp = spacy.load("en_core_web_sm")
         except OSError:
-            print(f"Modelo {model_name} no encontrado. Descargando...")
-            download(model_name)
-            self.nlp = spacy.load(model_name)
-            
+            raise RuntimeError(
+                "No se encontró el modelo en_core_web_sm. "
+                "Instálalo durante el build con: "
+                "python -m spacy download en_core_web_sm"
+            )
+
         self.sia = SentimentIntensityAnalyzer()
-        self.terminos_odio = ["stupid", "useless", "disease", "parasite"]
 
-    def analizar(self, texto):
+        # Lista inicial de términos ofensivos
+        self.terminos_odio = {
+            "stupid",
+            "useless",
+            "disease",
+            "parasite"
+        }
+
+
+    def analizar(self, texto: str) -> str:
+        """
+        Analiza una frase y devuelve una categoría.
+        """
+
+        if not texto or not texto.strip():
+            return "Neutral"
+
+
+        texto = texto.strip()
+
         doc = self.nlp(texto)
-        score = self.sia.polarity_scores(texto)
-        
-        ataque = False
+
+        sentimiento = self.sia.polarity_scores(texto)
+
+
+        ataque = self._detectar_ataque(doc)
+
+        ironia = self._detectar_ironia(
+            texto,
+            sentimiento
+        )
+
+
+        if ataque and ironia:
+            return "Sarcastic Hate Speech"
+
+        if ataque:
+            return "Direct Hate Speech"
+
+        if ironia:
+            return "Sarcastic/Ironic (Non-Hateful)"
+
+        return "Neutral"
+
+
+
+    def _detectar_ataque(self, doc) -> bool:
+        """
+        Detecta estructuras del tipo:
+
+        "They are parasites"
+        "You are useless"
+        """
+
         for token in doc:
+
+            # sujeto + verbo ser
             if token.dep_ == "nsubj" and token.head.lemma_ == "be":
+
                 for child in token.head.children:
-                    if child.dep_ in ["acomp", "attr"] and child.text.lower() in self.terminos_odio:
-                        ataque = True
-        
-        es_ironico = (score['pos'] > 0.3) and any(w in texto.lower() for w in ["always", "oh", "sure"])
-        
-        if ataque and es_ironico: return "Sarcastic Hate Speech"
-        if ataque: return "Direct Hate Speech"
-        if es_ironico: return "Sarcastic/Ironic (Non-Hateful)"
-        return "Neutral"
+
+                    if (
+                        child.dep_ in ["acomp", "attr"]
+                        and child.text.lower() in self.terminos_odio
+                    ):
+                        return True
+
+        return False
+
+
+
+    def _detectar_ironia(
+        self,
+        texto: str,
+        sentimiento: dict
+    ) -> bool:
+        """
+        Heurística simple de ironía.
+
+        Ejemplos:
+        "Oh sure, they are always perfect"
+        """
+
+        marcadores_ironia = [
+            "always",
+            "oh",
+            "sure",
+            "yeah",
+            "right",
+            "obviously"
+        ]
+
+
+        tiene_marcador = any(
+            palabra in texto.lower()
+            for palabra in marcadores_ironia
+        )
+
+
+        sentimiento_positivo = (
+            sentimiento["pos"] > 0.3
+        )
+
+
+        return (
+            tiene_marcador
+            and sentimiento_positivo
+        )