mirror of
https://github.com/NikolaiT/se-scraper.git
synced 2025-08-13 23:38:55 +02:00
before_keyword_scraped() hook supported
This commit is contained in:
4
run.js
4
run.js
@ -10,7 +10,7 @@ let config = {
|
||||
write_meta_data: false,
|
||||
// how long to sleep between requests. a random sleep interval within the range [a,b]
|
||||
// is drawn before every request. empty string for no sleeping.
|
||||
sleep_range: '',
|
||||
sleep_range: '[1,1]',
|
||||
// which search engine to scrape
|
||||
search_engine: 'google',
|
||||
// whether debug information should be printed
|
||||
@ -20,7 +20,7 @@ let config = {
|
||||
// this output is informational
|
||||
verbose: false,
|
||||
// an array of keywords to scrape
|
||||
keywords: ['incolumitas.com news', ],
|
||||
keywords: ['trump', 'chief'],
|
||||
// alternatively you can specify a keyword_file. this overwrites the keywords array
|
||||
keyword_file: '',
|
||||
// whether to start the browser in headless mode
|
||||
|
Reference in New Issue
Block a user