Rekomendasi Microsoft News

Buku catatan demo untuk mengakses data MIND di Azure

Buku catatan ini menyediakan contoh mengakses data MIND dari penyimpanan blob di Azure.

Data MIND disimpan di pusat data AS Barat/Timur, sehingga buku catatan ini akan berjalan dengan lebih efisien pada komputasi Azure yang terletak di AS Barat/Timur.

Impor dan lingkungan

import os
import tempfile
import shutil
import urllib
import zipfile
import pandas as pd

# Temporary folder for data we need during execution of this notebook (we'll clean up
# at the end, we promise)
temp_dir = os.path.join(tempfile.gettempdir(), 'mind')
os.makedirs(temp_dir, exist_ok=True)

# The dataset is split into training and validation set, each with a large and small version.
# The format of the four files are the same.
# For demonstration purpose, we will use small version validation set only.
base_url = 'https://mind201910small.blob.core.windows.net/release'
training_small_url = f'{base_url}/MINDsmall_train.zip'
validation_small_url = f'{base_url}/MINDsmall_dev.zip'
training_large_url = f'{base_url}/MINDlarge_train.zip'
validation_large_url = f'{base_url}/MINDlarge_dev.zip'

Fungsi

def download_url(url,
                 destination_filename=None,
                 progress_updater=None,
                 force_download=False,
                 verbose=True):
    """
    Download a URL to a temporary file
    """
    if not verbose:
        progress_updater = None
    # This is not intended to guarantee uniqueness, we just know it happens to guarantee
    # uniqueness for this application.
    if destination_filename is None:
        url_as_filename = url.replace('://', '_').replace('/', '_')
        destination_filename = \
            os.path.join(temp_dir,url_as_filename)
    if (not force_download) and (os.path.isfile(destination_filename)):
        if verbose:
            print('Bypassing download of already-downloaded file {}'.format(
                os.path.basename(url)))
        return destination_filename
    if verbose:
        print('Downloading file {} to {}'.format(os.path.basename(url),
                                                 destination_filename),
              end='')
    urllib.request.urlretrieve(url, destination_filename, progress_updater)
    assert (os.path.isfile(destination_filename))
    nBytes = os.path.getsize(destination_filename)
    if verbose:
        print('...done, {} bytes.'.format(nBytes))
    return destination_filename

Unduh dan ekstrak file

# For demonstration purpose, we will use small version validation set only.
# This file is about 30MB.
zip_path = download_url(validation_small_url, verbose=True)
with zipfile.ZipFile(zip_path, 'r') as zip_ref:
    zip_ref.extractall(temp_dir)

os.listdir(temp_dir)

Baca file dengan pandas

# The behaviors.tsv file contains the impression logs and users' news click histories. 
# It has 5 columns divided by the tab symbol:
# - Impression ID. The ID of an impression.
# - User ID. The anonymous ID of a user.
# - Time. The impression time with format "MM/DD/YYYY HH:MM:SS AM/PM".
# - History. The news click history (ID list of clicked news) of this user before this impression.
# - Impressions. List of news displayed in this impression and user's click behaviors on them (1 for click and 0 for non-click).
behaviors_path = os.path.join(temp_dir, 'behaviors.tsv')
pd.read_table(
    behaviors_path,
    header=None,
    names=['impression_id', 'user_id', 'time', 'history', 'impressions'])

# The news.tsv file contains the detailed information of news articles involved in the behaviors.tsv file.
# It has 7 columns, which are divided by the tab symbol:
# - News ID
# - Category
# - Subcategory
# - Title
# - Abstract
# - URL
# - Title Entities (entities contained in the title of this news)
# - Abstract Entities (entities contained in the abstract of this news)
news_path = os.path.join(temp_dir, 'news.tsv')
pd.read_table(news_path,
              header=None,
              names=[
                  'id', 'category', 'subcategory', 'title', 'abstract', 'url',
                  'title_entities', 'abstract_entities'
              ])

# The entity_embedding.vec file contains the 100-dimensional embeddings
# of the entities learned from the subgraph by TransE method.
# The first column is the ID of entity, and the other columns are the embedding vector values.
entity_embedding_path = os.path.join(temp_dir, 'entity_embedding.vec')
entity_embedding = pd.read_table(entity_embedding_path, header=None)
entity_embedding['vector'] = entity_embedding.iloc[:, 1:101].values.tolist()
entity_embedding = entity_embedding[[0,
                                     'vector']].rename(columns={0: "entity"})
entity_embedding

# The relation_embedding.vec file contains the 100-dimensional embeddings
# of the relations learned from the subgraph by TransE method.
# The first column is the ID of relation, and the other columns are the embedding vector values.
relation_embedding_path = os.path.join(temp_dir, 'relation_embedding.vec')
relation_embedding = pd.read_table(relation_embedding_path, header=None)
relation_embedding['vector'] = relation_embedding.iloc[:,
                                                       1:101].values.tolist()
relation_embedding = relation_embedding[[0, 'vector'
                                         ]].rename(columns={0: "relation"})
relation_embedding

Bersihkan file sementara

shutil.rmtree(temp_dir)