app.py

# ================== imports ==================
import gradio as gr
import pandas as pd
import numpy as np
import os, tempfile, re
from sentence_transformers import SentenceTransformer, util
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
import gdown

# ================== إعدادات ==================
BOOKS_FILE = "book.xlsx"
THESES_FILE = "theses.xlsx"

DRIVE_BOOKS_ID = "1IVu5M_zHWg3wo-cbQm3FWhmDIafUSCVP"
DRIVE_THESES_ID = "1hvZIGFG6h0kQY32bIuEWGRdXTZp84vhk"

TOP_K = 20
TFIDF_TOP_N = 100
TFIDF_FALLBACK_MIN = 30

# ================== نموذج Semantic ==================
model = SentenceTransformer("all-MiniLM-L6-v2")

# ================== تحميل الملفات ==================
def download_from_drive(file_id, output):
    if not os.path.exists(output):
        url = f"https://drive.google.com/uc?id={file_id}"
        gdown.download(url, output, quiet=True)

download_from_drive(DRIVE_BOOKS_ID, BOOKS_FILE)
download_from_drive(DRIVE_THESES_ID, THESES_FILE)

# ================== تحميل البيانات ==================
books_df = pd.read_excel(BOOKS_FILE).fillna("")
theses_df = pd.read_excel(THESES_FILE).fillna("")

books_df["المصدر"] = "كتاب"
theses_df["المصدر"] = "رسالة"

for df in [books_df, theses_df]:
    if "الموقع على الرف" in df.columns:
        df["الموقع على الرف"] = (
            df["الموقع على الرف"]
            .astype(str)
            .str.strip()
            .replace(["nan", "NaN", "None", ""], "-")
        )

library_df = pd.concat([books_df, theses_df], ignore_index=True)

# ================== الأعمدة الدلالية ==================
def detect_semantic_columns(df):
    exclude = ["م", "سنة", "تاريخ", "رقم", "كود", "الموقع", "الرف"]
    include = ["عنوان", "العنوان", "مؤلف", "المؤلف", "موضوع", "ملخص", "وصف", "كلمات"]

    cols = []
    for c in df.columns:
        s = str(c)
        if any(x in s for x in exclude):
            continue
        if any(x in s for x in include):
            cols.append(c)

    if not cols:
        cols = df.select_dtypes(include="object").columns.tolist()

    return cols

semantic_cols = detect_semantic_columns(library_df)

# ================== Normalize ==================
def normalize_text(text):
    text = str(text).lower()
    text = re.sub(r"[^\w\s]", " ", text)
    text = text.replace("أ", "ا").replace("إ", "ا").replace("آ", "ا")
    text = text.replace("ى", "ي").replace("ة", "ه")
    return re.sub(r"\s+", " ", text).strip()

def row_to_text(row):
    return normalize_text(
        " ".join(str(row[c]) for c in semantic_cols if row[c])
    )

# ================== Embeddings ==================
embeddings = model.encode(
    library_df.apply(row_to_text, axis=1).tolist(),
    convert_to_numpy=True,
    normalize_embeddings=True,
    show_progress_bar=True
)
library_df["embedding"] = list(embeddings)

# ================== TF-IDF ==================
def tfidf_prefilter(df, query):
    texts = df.apply(row_to_text, axis=1).tolist()
    vec = TfidfVectorizer(ngram_range=(1, 2), max_features=20000)
    mat = vec.fit_transform(texts)
    q_vec = vec.transform([normalize_text(query)])

    scores = cosine_similarity(q_vec, mat)[0]
    df = df.copy()
    df["tfidf_score"] = scores

    return df.sort_values("tfidf_score", ascending=False).head(TFIDF_TOP_N)

# ================== CSS ==================
CUSTOM_CSS = """
<style>
.card{border:1px solid #ddd;border-radius:8px;padding:10px;margin:12px 0;background:#fafafa;}
.card table{width:100%;border-collapse:collapse;direction:rtl;}
.card td{border:1px solid #ccc;padding:8px;text-align:center;vertical-align:middle;font-size:15px;}
.card td.label{width:30%;font-weight:bold;}
.row-title td{background:#e3f2fd;color:#0d47a1;font-weight:bold;}
.row-author td{background:#fce4ec;color:#880e4f;}
.row-year td{background:#e8f5e9;color:#1b5e20;}
.row-shelf td{background:#fff8e1;color:#e65100;font-weight:bold;font-size:16px;}
</style>
"""

def safe(v):
    return "-" if pd.isna(v) or str(v).strip() == "" else str(v)

def results_to_html(df):
    if df.empty:
        return "<p>❌ لا توجد نتائج</p>"

    html = ""
    for _, r in df.iterrows():
        html += f"""
        <div class="card">
        <table>
            <tr class="row-title">
                <td class="label">العنوان</td>
                <td>{safe(r.get("العنوان"))}</td>
            </tr>
            <tr class="row-author">
                <td class="label">المؤلف</td>
                <td>{safe(r.get("المؤلف"))}</td>
            </tr>
            <tr class="row-year">
                <td class="label">سنة النشر</td>
                <td>{safe(r.get("سنة النشر"))}</td>
            </tr>
            <tr class="row-shelf">
                <td class="label">الموقع على الرف</td>
                <td>{safe(r.get("الموقع على الرف"))}</td>
            </tr>
        </table>
        </div>
        """

    return CUSTOM_CSS + html

# ================== البحث ==================
def local_search_df(query, mode, source_filter):
    if not query.strip():
        return "<p>⚠️ اكتب كلمة للبحث</p>", pd.DataFrame()

    df = library_df.copy()

    if source_filter != "الكل":
        df = df[df["المصدر"] == source_filter]

    if df.empty:
        return "<p>❌ لا توجد بيانات</p>", pd.DataFrame()

    if mode == "نصي":
        mask = (
            df["العنوان"].str.contains(query, case=False, na=False) |
            df["المؤلف"].str.contains(query, case=False, na=False)
        )
        df = df[mask].head(TOP_K)
    else:
        df_pref = tfidf_prefilter(df, query)
        if len(df_pref) < TFIDF_FALLBACK_MIN:
            df_pref = df

        q_emb = model.encode([normalize_text(query)], normalize_embeddings=True)
        mat = np.vstack(df_pref["embedding"].values)
        scores = util.cos_sim(q_emb, mat)[0].cpu().numpy()

        df_pref["score"] = scores
        df = df_pref.sort_values("score", ascending=False).head(TOP_K)

    return results_to_html(df), df

# ================== حفظ النتائج ==================
def save_to_excel(df):
    tmp = tempfile.NamedTemporaryFile(delete=False, suffix=".xlsx")
    df.drop(columns=["embedding", "score", "tfidf_score"], errors="ignore") \
      .to_excel(tmp.name, index=False)
    return tmp.name

# ================== واجهة Gradio ==================
with gr.Blocks(title="نظام البحث الذكي بالمكتبة") as app:
    gr.Markdown("## 🔍 البحث الذكي في مقتنيات المكتبة")

    query = gr.Textbox(label="كلمة أو موضوع البحث")
    mode = gr.Radio(["نصي", "دلالي (Semantic)"], value="نصي", label="نوع البحث")
    source_filter = gr.Radio(["الكل", "كتاب", "رسالة"], value="الكل", label="المصدر")

    btn = gr.Button("🔎 بحث")
    out_html = gr.HTML()
    df_state = gr.State()
    file_out = gr.File(label="⬇️ تحميل النتائج")

    btn.click(local_search_df, [query, mode, source_filter], [out_html, df_state])
    gr.Button("📥 حفظ النتائج").click(save_to_excel, df_state, file_out)

app.launch()