I've finally built a first iteration of a job extraction pipeline in my job-advert-analysis repository. There's nothing in there that I haven't written about, but it's simply doing the work to bring it all together. I'm really happy to have a full pipeline that extracts lots of interesting features to analyse, and is easy to extend.
The data is taken from Common Crawl's scans of the web. We have a set of predefined URL patterns to search on Common Crawl that correspond to different job ads, that were found through various means. These are searched for in the Common Crawl capture index to find where all the corresponding pages are. Then all the corresponding pages are downlaoded from S3 as WARC.
The pages are then converted to JSON Lines for each source, using a custom extractor for each source capturing the semi-structured data. These are then normalised (e.g. HTML to text, extract salary, normalise location) and output as a feather dataframe.
With just two crawls (which happen about once a month) I've got over 13,000 job ads. Now I've got all this together I'm going to start analysing the results; and probably finding some bugs in the pipeline!