Web Scraping Risk Factors from 10-K EDGAR

I spent a lot of time trying to develop a methodology using REGEX and had some limited success. The problem is that the underlying XML submitted to the SEC does not strictly adhere to standard, and many reports deviate from the report format used. Sometimes they'll use capital case, title case, or use different combinations of letters and numbers to delineate sections. Sometimes they'll include introductory paragraphs to provide additional context to the risks they're about to list. There are so many random factors that interfere with the ability to establish any kind of pattern to the document structure that it's currently more efficient to parse these by human than machine. But, there are thousands upon thousands of documents, making this a very tedious, expensive and drawn out process. One method that might be useful is Amazon's Mechanical Turk, but that would still likely require a lot of upfront development time and could lead to cost restrictions unless the project is well funded.