This Python script parses the https://www.justice.gov/news site for links to all releases which are parsed with BeautifulSoup and stored as JSON entities, and combined into a single JSON file. For each Press Release, the following is captured:
- Press Release Number (Can be missing)
- Title
- Contents
- Publish Date
- Topics (If any are given)
- Components (Related agencies / deparments, if any)
As of 2018-07-28, this script has downloaded 13,087 Press Releases while the DOJ website lists 13,129. The difference comes from some pages which exist, but do not contain any links. For example, at the time of scrape, page 311 did not contain any links, but there were links on page 310 and 312. The data spans through January 5th, 2009.
You can simply query the dataset from BigQuery at https://bigquery.cloud.google.com/table/jbencina-144002:doj.press_releases or download from Kaggle at https://www.kaggle.com/jbencina/department-of-justice-20092018-press-releases
You could also run the scraper, however there is a 1.5 second delay between scrape requests. That means (1.5 sec x 599 links) + (1.5 sec x 13,087 releases) = at least 5 hours, 42 minutes.
Below is an example BQ query which flattens the topics and components since they are stored as string arrays.
Query all records
#standardSQL
SELECT
id,
title,
contents,
date,
ARRAY_TO_STRING(topics,",") topics,
ARRAY_TO_STRING(components,",") components
FROM
`jbencina-144002.doj.press_releases`
scraper.py - Main file to run which will scrape and download content
definitions.py - Cleaner way to store the BeautifulSoup attributes
Example Query.ipynb - Shows example of how to query BQ from Pandas
Keras Classifier Model.ipynb - Shows example of RNN model with detail here https://medium.com/jbencina/scraping-text-data-into-bigquery-and-building-a-keras-rnn-classifier-model-from-doj-press-releases-7899e520a1cf
- Ensure
BeautifulSoupandrequestslibraries are installed - Run
scraper.py