Each scraped article has the following fields:
- title: Title of the article
- datetime: Publication date
- content: Full content (text format)
- link: URL where the article was published
- keyword: Google News keyword used to find this article
No upper bound of course but it should be in the range 100,000 articles per day when scraping 24/7 with VPN enabled.
git clone git@github.com:philipperemy/google-news-scraper.git gns
cd gns
sudo pip install -r requirements.txt
python main_no_vpn.py # for VPN support, scroll down!
Article 1
{
"content": "(本文中の野村証券 [...] 生命経済研の熊野英生氏は指摘。 記事の全文 \n保護主義を根拠とする円高説を信じ込むのは禁物であり、実際は米貿易赤字縮小と円安が進むかもしれないとＢＢＨの村田雅志氏は指摘。 記事の全文 \n",
"datetime": "2015/11/03",
"keyword": "米国の銀行業務",
"link": "http://jp.reuters.com/article/idJPL3N12Y5QX20151104",
"title": "再送-インタビュー：運用高度化、ＰＥやハイイールド債増やす＝長門・ゆうちょ銀社長"
}
Article 2
{
"content": "記事保存 有料会員の方のみご利用になれます。[...] 詳しくは、こちら 電子版トップ速報トップ アルゼンチン、ドル、通貨ペソ、外貨取引 来春の新入社員を募集 記者など４職種 【週末新紙面】宅配＋電子版お試し実施中！ 天気 プレスリリース検索 アカウント一覧 訂正・おわび",
"datetime": "2015/12/17",
"keyword": "アルゼンチン",
"link": "http://www.nikkei.com/article/DGXLASGM18H1B_Y5A211C1EAF000/",
"title": "アルゼンチンの通貨ペソ、大幅下落 対ドルで36％安"
}
NOTE: The field content was truncated for improving the readibility.
SLEEP_TIME_EVERY_TEN_ARTICLES_IN_SECONDS: Sleep time before two calls to Google News. On average 10 articles are fetched per call. Default value is 1 second.ARTICLE_COUNT_LIMIT_PER_KEYWORD: Maximum number of articles fetched for one keyword. Default value is 300. I tried it up to 600 and it worked.RUN_POST_PROCESSING: Post processing means opening the URL of the article and extracting the content. For maximum efficiency, we first scrape all the available tuples (title, datetime, url) on Google.com. Then, from the collected URLs, we fetch the content. This two-step procedure is empirically more efficient. Run firstRUN_POST_PROCESSINGwith a value of 0. Then, run it a second time withRUN_POST_PROCESSINGset to 1. All the Google data scraped is persisted so no problem!LINKS_POST_PROCESSING_CLEAN_HTML_RATIO_LETTERS_LENGTH: Technical parameter for the post processing. Apply to Japanese only. We are interesting in dropping the english sentences from the Japanese articles. Default is 0.33.LINKS_POST_PROCESSING_NUM_THREADS: Number of threads to use when doing this post processing task. Default is 8.
Scraping Google News usually results in a ban for a few hours. Using a VPN with dynamic IP fetching is a way to overcome this problem.
In my case, I subscribed to this VPN: https://www.expressvpn.com/.
I provide a python binding for this VPN here: https://github.com/philipperemy/expressvpn-python.
Run those commands in Ubuntu 64 bits to configure the VPN with the Google News Scraper project:
git clone git@github.com:philipperemy/expressvpn-python.git evpn
cd evpn
sudo dpkg -i expressvpn_1.2.0_amd64.deb # will install the binaries provided by ExpressVPN
sudo pip install . # will install it as a package
Also make sure that:
- you can run
expressvpnin your terminal. - ExpressVPN is properly configured:
- you get
expressvpn-python (x.y)wherex.yis the version, when you runpip list | grep "expressvpn-python"
Once you have all of that, simply run:
# I advise you to set RUN_POST_PROCESSING=0 in conf.json, run main.py.
# It will gather all the Google data in data/. Then stop the script.
# Change the parameter RUN_POST_PROCESSING=1 and run it again.
# The script will read from data/ and start fetching the content of the links.
python main.py
Every time the script detects that Google has banned you, it will request the VPN to get a fresh new IP and will resume.
- Why didn't you use the RSS feed provided by Google News? It does not exist for Japanese!
- What is the best way to use this scraper? If you want to scrape a lot of data, I highly recommend you to subscribe to a VPN, preferably ExpressVPN (I implemented the VPN wrapper and the interaction with this scraper).