Create app.py

app.py

@@ -0,0 +1,117 @@
+import re
+import gradio as gr
+from transformers import pipeline
+
+# =========================
+# LOAD MODEL (ONCE)
+# =========================
+classifier = pipeline(
+    "zero-shot-classification",
+    model="joeddav/xlm-roberta-large-xnli",
+    device=-1
+)
+
+# =========================
+# SEMANTIC EVENT DEFINITIONS
+# =========================
+SEMANTIC_EVENT = {
+    "sadisme": [
+        "kekerasan ekstrem yang menyebabkan penderitaan fisik parah pada manusia",
+        "penyiksaan tubuh manusia hingga kematian",
+        "pengorbanan manusia secara kejam"
+    ],
+    "kekerasan_fisik": [
+        "tindakan fisik yang melukai tubuh manusia",
+        "serangan fisik yang menyebabkan cedera atau kematian"
+    ],
+    "kekerasan_verbal": [
+        "ancaman serius terhadap seseorang",
+        "penghinaan keras dan intimidasi verbal"
+    ],
+    "seksual": [
+        "aktivitas seksual eksplisit antara manusia",
+        "tindakan seksual yang digambarkan secara jelas"
+    ],
+    "perjudian": [
+        "aktivitas perjudian dengan taruhan uang atau barang"
+    ],
+    "narkoba": [
+        "penggunaan atau peredaran narkotika atau obat terlarang"
+    ]
+}
+
+# =========================
+# UTILITIES
+# =========================
+def split_kalimat(text):
+    return [
+        s.strip()
+        for s in re.split(r'(?<=[.!?])\s+', text)
+        if len(s.strip()) >= 30
+    ]
+
+def sliding_window(kalimat, window=4):
+    return [
+        " ".join(kalimat[i:i+window])
+        for i in range(len(kalimat) - window + 1)
+    ]
+
+def rating_usia(kategori):
+    if "sadisme" in kategori:
+        return 21
+    if any(k in kategori for k in ["seksual", "perjudian", "narkoba"]):
+        return 17
+    if any(k in kategori for k in ["kekerasan_fisik", "kekerasan_verbal"]):
+        return 13
+    return 0
+
+# =========================
+# CORE ANALYSIS FUNCTION
+# =========================
+def analyze_story(judul, isi):
+    if not isi or len(isi) < 100:
+        return {
+            "categories": [],
+            "age_rating": 0
+        }
+
+    kalimat = split_kalimat(isi)
+    windows = sliding_window(kalimat)
+
+    detected = set()
+
+    for window in windows:
+        for kategori, desc in SEMANTIC_EVENT.items():
+            result = classifier(
+                window,
+                desc,
+                hypothesis_template="Teks ini mengandung {}."
+            )
+            if max(result["scores"]) >= 0.80:
+                detected.add(kategori)
+
+    return {
+        "categories": sorted(list(detected)),
+        "age_rating": rating_usia(detected)
+    }
+
+# =========================
+# GRADIO INTERFACE
+# =========================
+def api_wrapper(judul, isi):
+    result = analyze_story(judul, isi)
+    return result
+
+demo = gr.Interface(
+    fn=api_wrapper,
+    inputs=[
+        gr.Textbox(label="Judul", lines=1),
+        gr.Textbox(label="Isi Cerita", lines=12)
+    ],
+    outputs=gr.JSON(label="Hasil Analisis"),
+    title="READEAR — Semantic Content Rating",
+    description="Analisis konten cerita berbasis semantic (zero-shot NLP)"
+)
+
+if __name__ == "__main__":
+    demo.launch()