app.py

from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
from transformers import pipeline
from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity
import pandas as pd
import os

app = FastAPI(title="MoodRead API", version="2.0.0")

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["*"],
    allow_headers=["*"],
)

# ── 1. Load model emosi RoBERTa ──────────────────────────────────────────────
MODEL_NAME = "StevenLimcorn/indonesian-roberta-base-emotion-classifier"
print("Memuat model emosi RoBERTa...")
emotion_pipeline = pipeline(
    task="text-classification",
    model=MODEL_NAME,
    device=-1   # CPU
)
print("✓ Model emosi siap!")

# ── 2. Load Sentence Transformer (Dania) ─────────────────────────────────────
ST_MODEL = "paraphrase-multilingual-MiniLM-L12-v2"
print("Memuat Sentence Transformer...")
sentence_model = SentenceTransformer(ST_MODEL)
print("✓ Sentence Transformer siap!")

# ── 3. Load dataset buku ─────────────────────────────────────────────────────
DATASET_PATH = "books_clean.csv"
if os.path.exists(DATASET_PATH):
    books_df = pd.read_csv(DATASET_PATH, encoding="utf-8")
    books_df["genre"] = books_df["genre"].str.lower().str.strip()
    print(f"✓ Dataset buku dimuat: {len(books_df)} buku")
else:
    books_df = pd.DataFrame()
    print("⚠️  books_clean.csv tidak ditemukan!")

# ── Schema ───────────────────────────────────────────────────────────────────
class TextInput(BaseModel):
    text: str
    top_k: int = 5

# ── Fungsi Yuri: Hybrid Contextual Genre Mapping ─────────────────────────────
def get_recommended_genres(emotion: str, text_user: str) -> list[str]:
    """
    Logika hybrid: kombinasi hasil model AI + aturan kata kunci kontekstual.
    """
    teks_bersih = text_user.lower().strip()
    emosi_bersih = emotion.lower().strip()

    kata_kunci_karir = [
        'presentasi', 'ujian', 'sidang', 'kerja', 'tugas', 'dosen',
        'kuliah', 'gagal', 'bos', 'kantor', 'capek', 'skripsi', 'kelulusan'
    ]
    kata_kunci_sosial = [
        'sahabat', 'cinta', 'putus', 'dikhianati', 'pacar', 'mantan',
        'teman', 'kucing', 'kehilangan', 'ditipu', 'rekan bisnis'
    ]

    if emosi_bersih in ['happy', 'joy']:
        return ['pengembangan_diri', 'novel']

    elif emosi_bersih == 'love':
        return ['romance']

    elif emosi_bersih == 'fear':
        if any(k in teks_bersih for k in kata_kunci_karir):
            return ['pengembangan_diri']
        else:
            return ['novel', 'horor']

    elif emosi_bersih == 'sadness':
        if any(k in teks_bersih for k in ['marah', 'kesal', 'ditipu', 'ketidakadilan', 'geram']):
            return ['novel', 'thriller']
        elif any(k in teks_bersih for k in kata_kunci_sosial):
            return ['romance']
        elif any(k in teks_bersih for k in kata_kunci_karir):
            return ['pengembangan_diri']
        else:
            return ['fiksi_sastra']

    elif emosi_bersih == 'anger':
        if any(k in teks_bersih for k in kata_kunci_karir):
            return ['cerpen']
        else:
            return ['novel', 'thriller']

    return ['novel', 'fiksi_sastra']


# ── Fungsi Yuri: Deteksi emosi mentah ────────────────────────────────────────
def detect_emotion(text: str, top_k: int = 3) -> dict:
    if not text.strip():
        return {
            'primary_emotion': 'neutral',
            'primary_score': 0.0,
            'top_emotions': [],
        }
    raw = emotion_pipeline(text, top_k=top_k)
    return {
        'primary_emotion': raw[0]['label'].lower(),
        'primary_score': round(raw[0]['score'], 4),
        'top_emotions': [
            {'emotion': r['label'].lower(), 'score': round(r['score'], 4)}
            for r in raw
        ],
    }


# ── Fungsi Yuri: Full pipeline emosi + hybrid genre ──────────────────────────
def full_emotion_pipeline(text: str) -> dict:
    analisis = detect_emotion(text)
    genres = get_recommended_genres(analisis['primary_emotion'], text)
    return {
        'emotion': analisis['primary_emotion'],
        'confidence': analisis['primary_score'],
        'recommended_genres': genres,
        'all_emotions': analisis['top_emotions'],
    }


# ── Fungsi Dania: Rekomendasi buku via Sentence Transformer ──────────────────
def get_book_recommendations(user_text: str, target_genre: str, top_k: int = 5) -> list[dict]:
    if books_df.empty:
        return []

    df_filtered = books_df[
        books_df['genre'].str.contains(target_genre, case=False, na=False)
    ].copy()

    if df_filtered.empty:
        return []

    user_vector = sentence_model.encode([user_text])
    title_vectors = sentence_model.encode(df_filtered['judul'].tolist())
    similarities = cosine_similarity(user_vector, title_vectors)[0]
    df_filtered['similarity_score'] = similarities

    top_books = (
        df_filtered
        .sort_values('similarity_score', ascending=False)
        .head(top_k)
    )

    return top_books.drop(columns=['similarity_score']).to_dict('records')


# ── Fungsi utama: moodread_pipeline ──────────────────────────────────────────
def moodread_pipeline(teks_curhat: str, top_k: int = 5) -> dict:
    """
    Pipeline lengkap end-to-end:
    Tahap 1 → Deteksi emosi + hybrid genre mapping (Yuri)
    Tahap 2 → Rekomendasi buku via Sentence Transformer (Dania)
    """
    # Tahap 1: emosi & genre
    hasil_emosi = full_emotion_pipeline(teks_curhat)
    genre_utama = hasil_emosi['recommended_genres'][0]

    # Tahap 2: rekomendasi buku semantik
    buku = get_book_recommendations(teks_curhat, genre_utama, top_k=top_k)

    return {
        'emotion': hasil_emosi['emotion'],
        'confidence': hasil_emosi['confidence'],
        'all_emotions': hasil_emosi['all_emotions'],
        'recommended_genres': hasil_emosi['recommended_genres'],
        'book_recommendations': buku,
    }


# ── Endpoints ────────────────────────────────────────────────────────────────
@app.get("/")
def root():
    return {"message": "MoodRead API v2 aktif 🎉", "docs": "/docs"}

@app.get("/health")
def health():
    return {"status": "ok"}

@app.post("/detect-emotion")
def endpoint_detect(payload: TextInput):
    if not payload.text.strip():
        raise HTTPException(status_code=400, detail="Teks tidak boleh kosong.")

    # Panggil moodread_pipeline sebagai entry point utama
    hasil = moodread_pipeline(payload.text, top_k=payload.top_k)

    return {
        "primary_emotion":      hasil["emotion"],
        "primary_score":        hasil["confidence"],
        "top_emotions":         hasil["all_emotions"],
        "recommended_genres":   hasil["recommended_genres"],
        "book_recommendations": hasil["book_recommendations"],
    }

@app.get("/emotions")
def list_emotions():
    return {"emotions": ["happy", "joy", "love", "sadness", "anger", "fear", "surprise", "neutral"]}

@app.get("/genres")
def list_genres():
    if books_df.empty:
        return {"genres": []}
    return {"genres": books_df["genre"].unique().tolist()}