sentiment

app.py

@@ -1,7 +1,74 @@
 import gradio as gr
+from transformers import pipeline
+import math
 
-def greet(name):
-    return "Hello " + name + "!!"
+print("🎭 Loading sentiment analysis pipeline...")
+sentiment_analyzer = pipeline(
+    "sentiment-analysis",
+    model="cardiffnlp/twitter-roberta-base-sentiment-latest",
+    tokenizer="cardiffnlp/twitter-roberta-base-sentiment-latest",
+    return_all_scores=True,
+    truncation=True,
+    padding=True,
+    device_map="auto"
+)
 
-demo = gr.Interface(fn=greet, inputs="text", outputs="text")
-demo.launch()
+def _polarity_from_scores(scores):
+    # scores is a list like [{'label': 'negative', 'score': 0.01}, {'label':'neutral',...}, {'label':'positive',...}]
+    probs = {s["label"].lower(): float(s["score"]) for s in scores}
+    p_pos = probs.get("positive", 0.0)
+    p_neg = probs.get("negative", 0.0)
+    return p_pos - p_neg  # range roughly [-1, 1]
+
+def analyze_tone_and_bias(text, chunk_size=500, neutral_margin=0.1):
+    """
+    Analyze emotional tone and potential bias in sources.
+    Uses polarity = P(positive) - P(negative) and respects neutral.
+    """
+    # Make chunks
+    chunks = [text[i:i+chunk_size] for i in range(0, len(text), chunk_size)]
+    chunks = [c for c in chunks if len(c.strip()) > 10]
+    if not chunks:
+        return {"error": "Could not analyze sentiment"}
+
+    # Batch inference
+    batch_outputs = sentiment_analyzer(chunks)  # list of lists of dicts
+
+    # Compute per-chunk polarity
+    chunk_polarities = [_polarity_from_scores(scores) for scores in batch_outputs]
+
+    # Aggregate
+    avg_polarity = sum(chunk_polarities) / len(chunk_polarities)
+    if avg_polarity > neutral_margin:
+        overall = "POSITIVE"
+    elif avg_polarity < -neutral_margin:
+        overall = "NEGATIVE"
+    else:
+        overall = "NEUTRAL"
+
+    # Build a compact per-chunk view
+    # Use the model's top label for human-readable chunk summaries
+    tops = []
+    for scores in batch_outputs[:5]:
+        top = max(scores, key=lambda s: s["score"])
+        tops.append({"label": top["label"].upper(), "score": float(top["score"])})
+
+    return {
+        "overall_sentiment": overall,
+        "confidence": float(abs(avg_polarity)),
+        "chunk_analysis": tops
+    }
+
+print("✅ Sentiment analysis pipeline ready")
+
+def analyze_sentiment(text):
+  result = analyze_tone_and_bias(text)
+  return result['chunk_analysis'][0]
+
+demo = gr.Interface(fn=analyze_sentiment, inputs="textbox", outputs="textbox")
+
+test_txt = "Nærmeste samtlige biler og vogntog som passerte Ørskogfjellet da kontrollen pågikk ble stoppet. Hele 50 personer fra de ulike etatene var involvert i kontrollen."
+
+print(analyze_sentiment(test_txt))
+
+demo.launch(share=True)