Hi,

Dan Arkind from JobScore here. We have spent some time on this. Check out our stuff at www.jobscore.com or holler if you have more specific questions - dan@jobscore.com

The crux of what we do is a parse and clean pattern with a relatively straightforward resume fit scoring (or matching if you prefer, tomato, tomatoe).

You can enter weighted keyword terms (useful), but we also FILTER based on location, experience level (employers, job titles) and education level (schools, degrees). We've found this does a decent job of "wheat from the chaff" separation when you receive a lot of job applications.

The reality when you are hiring is that you definitely want to actually read and review ~ top 25% of inbound applications.

Finding matches in larger data sets (100M+ LinkedIn profiles) is obviously harder... but filtering one level down in structured data (degrees in education, job titles in employment) makes matches relatively precise.

There's a non-trivial challenge in normalizing job title information (javascript ninja vs. front end software engineer vs. awesome developer!) and degrees that could make matching more precise, but we aren't aware of anyone who has really taken this on (bueler?)

Variables that aren't typically captured (like salary) are really strong indicators of a match as well...

LinkedIn's (interesting) secret sauce here is relationship metadata (who you know) as a filter. LinkedIn also captures reputation metadata and the folks over at www.honestly.com are doubling down on this. We have an (undisclosed) secret sauce angle to matching at scale... but we'll need a heck of a lot more data for that to really matter...

In general though, you don't want or need an exact match (who you are going to hire), you just want what's close (who should I call).

Give jobscore a whirl & see if you have any additional input!

user:	danarkind
created:	August 26, 2009
karma:	3