VSS

Extract, transform, and load news stories from LexisNexis Bulk API dumps and from the LDC Gigaword corpora.

"Your [XML] seems less removed than hurled from you, and hurled with a velocity that lets you feel as through the [XML] is going to end up someplace so far away from you that it will have become an abstraction...a kind of existential-level [XML] treatment."

• extract_xml takes in one XML-format news article from the LexisNexis dump and returns a JSON object that looks like this:

    {'article_body': u'',
     'article_title': u'',
     'cities': [],
     'countries': [],
     'doc_id': u'',
     'news_source': u'',
     'position_section': u'',
     'publication_date': datetime.datetime(2015, 5, 18, 0, 0),
     'publication_date_raw': u'',
     'states': [],
     'word_count': u'',
     'id_type' : '',
    }

• gigaword_loader.py traverses a directory of LDC Gigaword documents in XML, transforms them, and loads them into a Mongo database. It takes as an argument the gigaword directory's path. The resulting documents look like this:

      { 'article_title' : article_title,
        'dateline' : dateline,
        'article_body' : article_body,
        'doc_id' : doc_id,
        'publication_date' : parsed_date,
        'news_source' : news_source,
        'language' : language,
        'doc_type' : doc_type,
        'word_count' : word_count
        }

Note: Python's built-in tools for logging and for multiprocessing don't work well together. I've defaulted to better logging because that saved more time in debugging than multiprocessing added, but all of the code for multiprocessing is built in and commented out. Switching to multiprocessing wouldn't be hard.

Note

The customer service I experienced from LexisNexis was terrible: rude, slow, and technically uninformed. I'd strongly consider other providers if they're available.