Update app.py

app.py

@@ -1,10 +1,9 @@
 import gradio as gr
 from transformers import pipeline
-import re
 
-# -----------------------------
-# Load classifier
-# -----------------------------
+# ----------------------------------
+# Load model from Hugging Face Hub
+# ----------------------------------
 MODEL_REPO = "MakD1227/afriberta-hsd-full"
 
 classifier = pipeline(
@@ -13,61 +12,47 @@ classifier = pipeline(
     tokenizer=MODEL_REPO
 )
 
-LABEL_MAP = {
-    "LABEL_0": "Free",
-    "LABEL_1": "Offensive",
-    "LABEL_2": "Hate"
-}
+# ----------------------------------
+# Prediction function
+# ----------------------------------
+def predict_speech(text):
+    results = classifier(text)
 
-# -----------------------------
-# Text segmentation (simple & robust)
-# -----------------------------
-def split_text(text):
-    # Split by punctuation and line breaks
-    segments = re.split(r'(?<=[።.!?])\s+|\n+', text)
-    return [seg.strip() for seg in segments if seg.strip()]
+    label_map = {
+        "LABEL_0": "Free (Neutral)",
+        "LABEL_1": "Offensive",
+        "LABEL_2": "Hate"
+    }
 
-# -----------------------------
-# Prediction with span labeling
-# -----------------------------
-def predict_with_spans(text):
-    segments = split_text(text)
-    highlighted = []
+    label = results[0]["label"]
+    score = results[0]["score"]
 
-    for seg in segments:
-        result = classifier(seg)[0]
-        label = LABEL_MAP[result["label"]]
-        highlighted.append((seg, label))
+    return label_map.get(label, label), f"{score * 100:.2f}%"
 
-    return highlighted
-
-# -----------------------------
+# ----------------------------------
 # Gradio Interface
-# -----------------------------
+# ----------------------------------
 demo = gr.Interface(
-    fn=predict_with_spans,
+    fn=predict_speech,
     inputs=gr.Textbox(
-        lines=4,
+        lines=2,
         label="Input Text",
-        placeholder="Enter mixed Amharic & Afan Oromo text..."
-    ),
-    outputs=gr.HighlightedText(
-        label="Detected Hate / Offensive / Free Segments",
-        color_map={
-            "Hate": "red",
-            "Offensive": "orange",
-            "Free": "green"
-        }
-    ),
-    title="Bilingual Hate Speech Detection (Amharic & Afan Oromo)",
-    description=(
-        "Fine-grained detection showing which portions of the text "
-        "are Hate, Offensive, or Free (supports code-mixed input)."
+        placeholder="Enter Amharic or Afan Oromo text..."
     ),
+    outputs=[
+        gr.Label(label="Classification"),
+        gr.Text(label="Confidence")
+    ],
+    title="Amharic & Afan Oromo Hate Speech Detector",
+    description="Classify text into Free, Offensive, or Hate Speech",
+    article="""
+    <p style='text-align:center;'>© 2025 Mequanent Degu Belete</p>
+    <p style='text-align:center;'>mekuanentde@gmail.com</p>
+    <p style='text-align:center;'>SNHCC, Academia Sinica, Taiwan</p>
+    """,
     examples=[
-        [
-            "ኢትዮጵያ ለዘላለም ትኑር haatee sali shamtuu situ nuu beekaa"
-        ]
+        ["ኢትዮጵያ ለዘላለም ትኑር"],
+        ["haatee sali shamtuu situ nuu beekaa waa ee baalee"]
     ]
 )