Update app.py

app.py

@@ -1,28 +1,55 @@
 from transformers import pipeline
 import gradio as gr
+import re
 
-# Load the model once (global)
-pipe = pipeline("text-classification", model="Titeiiko/OTIS-Official-Spam-Model")
+# ---------- Load models once (faster) ----------
+spam_pipe = pipeline("text-classification", model="Titeiiko/OTIS-Official-Spam-Model")
+zero_shot = pipeline("zero-shot-classification", model="facebook/bart-large-mnli")
 
-def is_spam(input_text: str) -> dict:
+# ---------- Helper ----------
+def is_gibberish(text: str) -> bool:
+    letters = len(re.findall(r"[a-zA-Z]", text))
+    total = len(text)
+    if total == 0:
+        return True
+    return letters / total < 0.6   # more than 40% non-letters → gibberish
+
+# ---------- Core Detection ----------
+def detect(text: str) -> dict:
     """
-    Returns {"spam": bool, "label": str, "score": float}
+    Returns a JSON-like dictionary with individual flags and final decision
     """
-    result = pipe(input_text)[0]
-    is_spam_flag = result["label"] != "LABEL_0"
+    # Ad/spam detection
+    spam_result = spam_pipe(text)[0]
+    spam_flag = spam_result["label"] != "LABEL_0"
+
+    # Off-topic / irrelevant detection
+    zero_result = zero_shot(text, candidate_labels=["relevant", "irrelevant"])
+    # zero-shot gives scores sorted by confidence
+    top_label = zero_result["labels"][0]
+    irrelevant_flag = top_label == "irrelevant"
+
+    # Gibberish detection
+    gibberish_flag = is_gibberish(text)
+
     return {
-        "spam": is_spam_flag,
-        "label": result["label"],
-        "score": float(result["score"])
+        "input": text,
+        "spam_flag": spam_flag,
+        "irrelevant_flag": irrelevant_flag,
+        "gibberish_flag": gibberish_flag,
+        "overall_flag": spam_flag or irrelevant_flag or gibberish_flag,
+        "spam_model_confidence": float(spam_result["score"]),
+        "zero_shot_top_label": top_label,
+        "zero_shot_confidence": float(zero_result["scores"][0])
     }
 
-# Gradio interface (for both API & UI)
+# ---------- Gradio Interface ----------
 demo = gr.Interface(
-    fn=is_spam,
-    inputs=gr.Textbox(label="Enter text"),
-    outputs=gr.JSON(label="Result"),
-    title="Spam Detector API",
-    description="Detect spam using Hugging Face Transformers."
+    fn=detect,
+    inputs=gr.Textbox(label="Enter text to classify"),
+    outputs=gr.JSON(label="Detection Result"),
+    title="Spam / Irrelevance / Gibberish Detector",
+    description="Combines spam detection, off-topic classification, and gibberish check."
 )
 
 if __name__ == "__main__":