scripts/add_wikidata.py

#!/usr/bin/env python3
"""
Add Wikidata ID and main category to UVW 2026 dataset.

Uses concurrent requests to Wikidata API for faster processing.

Fields added:
- wikidata_id: The Q-ID from Wikidata (e.g., Q1930)
- main_category: The P31 (instance of) property label in Vietnamese

API Reference:
- https://www.wikidata.org/w/api.php?action=help&modules=wbgetentities
"""

import json
import time
from concurrent.futures import ThreadPoolExecutor, as_completed
from pathlib import Path
from typing import Optional
import urllib.request
import urllib.parse
import urllib.error
import threading

from tqdm import tqdm

INPUT_PATH = Path(__file__).parent.parent / "data" / "processed" / "uvw_2026_quality.jsonl"
OUTPUT_PATH = Path(__file__).parent.parent / "data" / "processed" / "uvw_2026_wikidata.jsonl"
CACHE_PATH = Path(__file__).parent.parent / "data" / "processed" / "wikidata_cache.json"

# Wikidata API endpoint
WIKIDATA_API = "https://www.wikidata.org/w/api.php"

# Batch size for API requests (max 50 titles per request)
BATCH_SIZE = 50

# Number of concurrent workers
MAX_WORKERS = 8

# Request timeout
TIMEOUT = 60

# Thread-safe label cache
_label_cache: dict[str, Optional[str]] = {}
_label_cache_lock = threading.Lock()


def load_cache() -> dict:
    """Load cached Wikidata mappings."""
    if CACHE_PATH.exists():
        with open(CACHE_PATH, "r", encoding="utf-8") as f:
            return json.load(f)
    return {}


def save_cache(cache: dict):
    """Save Wikidata mappings to cache."""
    with open(CACHE_PATH, "w", encoding="utf-8") as f:
        json.dump(cache, f, ensure_ascii=False)


def fetch_entity_labels_batch(entity_ids: list[str]) -> dict[str, Optional[str]]:
    """Fetch Vietnamese/English labels for a batch of Wikidata entity IDs."""
    global _label_cache

    with _label_cache_lock:
        to_fetch = [eid for eid in entity_ids if eid not in _label_cache]

    if not to_fetch:
        with _label_cache_lock:
            return {eid: _label_cache.get(eid) for eid in entity_ids}

    ids_param = "|".join(to_fetch[:50])  # Max 50

    params = {
        "action": "wbgetentities",
        "ids": ids_param,
        "props": "labels",
        "languages": "vi|en",
        "format": "json",
    }

    url = f"{WIKIDATA_API}?{urllib.parse.urlencode(params)}"

    try:
        req = urllib.request.Request(
            url,
            headers={"User-Agent": "UVW2026-Dataset/1.0 (https://github.com/undertheseanlp)"}
        )
        with urllib.request.urlopen(req, timeout=TIMEOUT) as response:
            data = json.loads(response.read().decode("utf-8"))

        with _label_cache_lock:
            if "entities" in data:
                for entity_id, entity in data["entities"].items():
                    if "labels" in entity:
                        labels = entity["labels"]
                        if "vi" in labels:
                            _label_cache[entity_id] = labels["vi"]["value"]
                        elif "en" in labels:
                            _label_cache[entity_id] = labels["en"]["value"]
                        else:
                            _label_cache[entity_id] = None
                    else:
                        _label_cache[entity_id] = None

    except Exception:
        with _label_cache_lock:
            for eid in to_fetch[:50]:
                if eid not in _label_cache:
                    _label_cache[eid] = None

    with _label_cache_lock:
        return {eid: _label_cache.get(eid) for eid in entity_ids}


def fetch_wikidata_batch(titles: list[str], batch_id: int = 0) -> dict[str, dict]:
    """
    Fetch Wikidata entities for a batch of Vietnamese Wikipedia titles.

    Args:
        titles: List of Wikipedia article titles (max 50)
        batch_id: Batch identifier for error reporting

    Returns:
        Dict mapping title to {wikidata_id, main_category}
    """
    results = {}
    category_ids_to_fetch = []
    title_to_category_id = {}

    titles_param = "|".join(titles)

    params = {
        "action": "wbgetentities",
        "sites": "viwiki",
        "titles": titles_param,
        "props": "claims|sitelinks",
        "format": "json",
    }

    url = f"{WIKIDATA_API}?{urllib.parse.urlencode(params)}"

    try:
        req = urllib.request.Request(
            url,
            headers={"User-Agent": "UVW2026-Dataset/1.0 (https://github.com/undertheseanlp)"}
        )
        with urllib.request.urlopen(req, timeout=TIMEOUT) as response:
            data = json.loads(response.read().decode("utf-8"))

        if "entities" not in data:
            # Mark all as missing
            for title in titles:
                results[title] = {"wikidata_id": None, "main_category": None}
            return results

        # Track which titles we found
        found_titles = set()

        for entity_id, entity in data["entities"].items():
            if entity_id.startswith("-"):  # Missing entity
                continue

            # Get the title this entity corresponds to
            if "sitelinks" in entity and "viwiki" in entity["sitelinks"]:
                title = entity["sitelinks"]["viwiki"]["title"]
            else:
                continue

            found_titles.add(title)

            # Extract Wikidata ID
            wikidata_id = entity.get("id")

            # Extract P31 category ID
            category_id = None
            if "claims" in entity and "P31" in entity["claims"]:
                p31_claims = entity["claims"]["P31"]
                if p31_claims:
                    first_claim = p31_claims[0]
                    if "mainsnak" in first_claim:
                        mainsnak = first_claim["mainsnak"]
                        if mainsnak.get("datatype") == "wikibase-item":
                            datavalue = mainsnak.get("datavalue", {})
                            if datavalue.get("type") == "wikibase-entityid":
                                category_id = datavalue["value"].get("id")

            results[title] = {
                "wikidata_id": wikidata_id,
                "main_category": None,
            }

            if category_id:
                category_ids_to_fetch.append(category_id)
                title_to_category_id[title] = category_id

        # Mark missing titles
        for title in titles:
            if title not in found_titles:
                results[title] = {"wikidata_id": None, "main_category": None}

    except Exception as e:
        # On error, mark all as missing
        for title in titles:
            results[title] = {"wikidata_id": None, "main_category": None}
        return results

    # Fetch category labels
    if category_ids_to_fetch:
        unique_ids = list(set(category_ids_to_fetch))
        labels = fetch_entity_labels_batch(unique_ids)

        for title, category_id in title_to_category_id.items():
            if title in results:
                results[title]["main_category"] = labels.get(category_id)

    return results


def process_batch(args: tuple) -> tuple[int, dict[str, dict]]:
    """Process a single batch, returns (batch_id, results)."""
    batch_id, titles = args
    results = fetch_wikidata_batch(titles, batch_id)
    return batch_id, results


def count_lines(path: Path) -> int:
    """Count lines in a file without loading it all into memory."""
    count = 0
    with open(path, "r", encoding="utf-8") as f:
        for _ in f:
            count += 1
    return count


def iter_titles(path: Path):
    """Iterate over titles from JSONL file without loading full articles."""
    with open(path, "r", encoding="utf-8") as f:
        for line in f:
            # Extract title with minimal parsing
            article = json.loads(line)
            yield article["title"]


def iter_articles(path: Path):
    """Iterate over articles from JSONL file one at a time."""
    with open(path, "r", encoding="utf-8") as f:
        for line in f:
            yield json.loads(line)


def batch_iterator(iterable, batch_size: int):
    """Yield batches from an iterable without loading all into memory."""
    batch = []
    for item in iterable:
        batch.append(item)
        if len(batch) >= batch_size:
            yield batch
            batch = []
    if batch:
        yield batch


def main():
    """Add Wikidata information to dataset."""
    print("Adding Wikidata IDs and categories to UVW 2026 dataset...")
    print(f"Input: {INPUT_PATH}")
    print(f"Output: {OUTPUT_PATH}")
    print(f"Using {MAX_WORKERS} concurrent workers")

    # Load cache
    cache = load_cache()
    print(f"Loaded {len(cache):,} cached Wikidata mappings")

    # Count total articles (streaming, no memory)
    print("Counting articles...")
    total = count_lines(INPUT_PATH)
    print(f"Found {total:,} articles")

    # First pass: collect uncached titles (streaming, only store titles)
    print("Finding uncached titles...")
    uncached_titles = []
    for title in tqdm(iter_titles(INPUT_PATH), total=total, desc="Scanning titles"):
        if title not in cache:
            uncached_titles.append(title)
    print(f"Need to fetch {len(uncached_titles):,} titles from Wikidata API")

    # Fetch in batches using thread pool with controlled submission
    if uncached_titles:
        num_batches = (len(uncached_titles) + BATCH_SIZE - 1) // BATCH_SIZE
        print(f"Processing {num_batches:,} batches...")

        completed = 0
        last_save = 0

        # Process batches in chunks to limit memory from pending futures
        SUBMIT_CHUNK_SIZE = MAX_WORKERS * 10  # Limit pending futures

        with tqdm(total=num_batches, desc="Fetching Wikidata", smoothing=0.1) as pbar:
            with ThreadPoolExecutor(max_workers=MAX_WORKERS) as executor:
                batch_id = 0
                for chunk_start in range(0, len(uncached_titles), BATCH_SIZE * SUBMIT_CHUNK_SIZE):
                    chunk_end = min(chunk_start + BATCH_SIZE * SUBMIT_CHUNK_SIZE, len(uncached_titles))
                    chunk_titles = uncached_titles[chunk_start:chunk_end]

                    # Create batches for this chunk
                    title_batches = [chunk_titles[i:i + BATCH_SIZE]
                                     for i in range(0, len(chunk_titles), BATCH_SIZE)]

                    # Submit this chunk's batches
                    future_to_batch = {
                        executor.submit(process_batch, (batch_id + idx, titles_list)): batch_id + idx
                        for idx, titles_list in enumerate(title_batches)
                    }
                    batch_id += len(title_batches)

                    for future in as_completed(future_to_batch):
                        try:
                            _, results = future.result()
                            cache.update(results)
                            completed += 1
                            pbar.update(1)

                            # Save cache periodically
                            if completed - last_save >= 1000:
                                save_cache(cache)
                                last_save = completed

                        except Exception as e:
                            print(f"Batch error: {e}")

        # Clear uncached_titles to free memory before writing
        del uncached_titles

        # Final cache save
        save_cache(cache)
        print(f"Saved {len(cache):,} entries to cache")

    # Second pass: stream through articles and write output
    stats = {
        "with_wikidata_id": 0,
        "with_category": 0,
        "missing": 0,
    }
    category_counts: dict[str, int] = {}

    print("Writing output (streaming)...")
    with open(OUTPUT_PATH, "w", encoding="utf-8") as fout:
        for article in tqdm(iter_articles(INPUT_PATH), total=total, desc="Writing output"):
            title = article["title"]
            wikidata_info = cache.get(title, {})

            article["wikidata_id"] = wikidata_info.get("wikidata_id")
            article["main_category"] = wikidata_info.get("main_category")

            if article["wikidata_id"]:
                stats["with_wikidata_id"] += 1
            else:
                stats["missing"] += 1

            if article["main_category"]:
                stats["with_category"] += 1
                category = article["main_category"]
                category_counts[category] = category_counts.get(category, 0) + 1

            fout.write(json.dumps(article, ensure_ascii=False) + "\n")

    # Print statistics
    print("\nWikidata coverage:")
    print("-" * 50)
    print(f"  With Wikidata ID: {stats['with_wikidata_id']:>10,} ({stats['with_wikidata_id']/total*100:5.1f}%)")
    print(f"  With category:    {stats['with_category']:>10,} ({stats['with_category']/total*100:5.1f}%)")
    print(f"  Missing:          {stats['missing']:>10,} ({stats['missing']/total*100:5.1f}%)")

    print(f"\nTop 30 categories:")
    print("-" * 50)
    top_categories = sorted(category_counts.items(), key=lambda x: -x[1])[:30]
    for category, count in top_categories:
        pct = count / total * 100
        print(f"  {category:40s} {count:>8,} ({pct:5.2f}%)")

    print(f"\nTotal unique categories: {len(category_counts):,}")
    print(f"Total articles: {total:,}")
    print(f"Output saved to: {OUTPUT_PATH}")
    print(f"Cache saved to: {CACHE_PATH}")


if __name__ == "__main__":
    main()