app.py

# app.py
import gradio as gr
import yake
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import math

# --- Models / tools (küçük ve CPU-dostu)
MODEL_NAME = "google/flan-t5-small"
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME)

# YAKE keyword extractor (çok hafif)
def extract_keywords(text, lang="en", max_kw=8):
    kw_extractor = yake.KeywordExtractor(lan=lang, n=1, top=max_kw)
    kws = kw_extractor.extract_keywords(text)
    # returns list of keywords (sorted)
    return [kw for kw, score in kws]

# Basit SEO puanı hesaplama
def seo_score(title, description, keywords, tags):
    score = 0
    # title presence & length
    if title and title.strip():
        score += 15
        L = len(title)
        if 40 <= L <= 70:
            score += 20
        elif 30 <= L < 40 or 71 <= L <= 90:
            score += 10
    # keywords in title/description
    kw_in_title = sum(1 for k in keywords if k.lower() in (title or "").lower())
    kw_in_desc = sum(1 for k in keywords if k.lower() in (description or "").lower())
    score += min(20, kw_in_title * 10)
    score += min(15, kw_in_desc * 5)
    # description length
    dlen = len(description or "")
    if dlen >= 300:
        score += 20
    elif dlen >= 150:
        score += 10
    # tags
    if tags:
        if 3 <= len(tags) <= 15:
            score += 10
        elif len(tags) > 15:
            score += 5
    # normalize
    return min(100, score)

# Title & description generator (Flan-T5)
def gen_suggestions(main_text, keywords, max_titles=3):
    prompt = (
        "You are an assistant that generates catchy YouTube video titles and an SEO-optimized description.\n"
        f"Main content: {main_text}\n"
        f"Keywords: {', '.join(keywords)}\n"
        "Produce 3 short catchy titles (each <70 chars) and one SEO-friendly description (2 paragraphs). "
        "Return clearly, titles separated by '||' then '---' then the description."
    )
    inputs = tokenizer(prompt, return_tensors="pt")
    out = model.generate(**inputs, max_new_tokens=300)
    text = tokenizer.decode(out[0], skip_special_tokens=True)
    # Try to split results
    if "||" in text:
        parts = text.split("---")
        titles = parts[0].split("||")
        desc = parts[1].strip() if len(parts) > 1 else ""
    else:
        # fallback: guess
        lines = [l.strip() for l in text.split("\n") if l.strip()]
        titles = lines[:max_titles]
        desc = "\n".join(lines[max_titles:])
    titles = [t.strip() for t in titles if t.strip()]
    return titles[:max_titles], desc

# Gradio function
def analyze(title, description, lang_choice):
    text = (title or "") + "\n" + (description or "")
    lang = "tr" if lang_choice == "Türkçe" else "en"
    # extract keywords
    keywords = extract_keywords(text, lang=lang, max_kw=8)
    # generate suggestions
    gen_titles, gen_desc = gen_suggestions(text if text.strip() else "Short video about ...", keywords)
    # tags: use keywords as tags (shorten)
    tags = [k.replace(" ", "_") for k in keywords][:12]
    # score
    score = seo_score(title, description, keywords, tags)
    return {
        "SEO Skoru (0-100)": score,
        "Çıkarılan Anahtar Kelimeler": ", ".join(keywords),
        "Önerilen Başlıklar": "\n".join([f"- {t}" for t in gen_titles]),
        "Önerilen Açıklama": gen_desc,
        "Önerilen Etiketler (tags)": ", ".join(tags)
    }

# Gradio UI
with gr.Blocks() as demo:
    gr.Markdown("## YouTube SEO Asistanı — Basit & Ücretsiz (örnek)")
    with gr.Row():
        with gr.Column(scale=2):
            title_in = gr.Textbox(label="Mevcut başlık (isteğe bağlı)", lines=1, placeholder="Var olan videonuzun başlığını yazın")
            desc_in = gr.Textbox(label="Açıklama / Transcript / İçerik", lines=8, placeholder="Video açıklaması veya transkriptinizi buraya yapıştırın")
            lang = gr.Radio(choices=["Türkçe", "English"], value="Türkçe", label="Dil")
            btn = gr.Button("Analiz Et")
        with gr.Column(scale=1):
            score_out = gr.Label(num_top_classes=1, label="Sonuç")
            result_box = gr.JSON(label="Detaylı Öneriler")
    btn.click(fn=analyze, inputs=[title_in, desc_in, lang], outputs=[score_out, result_box])

if __name__ == "__main__":
    demo.launch()