Job Recommendation Engine

Question

My girlfriend has recently been struggling with finding a new job, so I thought I'd make a website to help her out. The basic idea is that she'll be shown a list of jobs, rate her interest, and then a hopefully interesting job will be emailed to her every day. So far I have the frontend worked out, but I'm struggling with the engine. I've looked into different kinds of recommendation engines, and it seems like a content based engine is the way to go since she will probably be the only user. With that said, feature selection is problematic because job categorizations kind of unrealistic. For instance, I could have job categories like "administrative" or "analytical", but those would be hard to define, especially if I was analyzing the job description for certain keywords.

Sorry about the wall of text. My basic question is how should I go about selecting features for this job recommendation engine?

Thanks for the help.

Answer 1

For a first pass at this problem, I suggest just use a simple document search/classifier feature set such as bag of words, or maybe tf-idf against each full job description, and see what results you can get from a basic classifier.

You can train the model based on your girlfriend short-listing or rejecting each item. These features can also be used more or less directly to find and order other documents by degree of similarity.

A bag of words model is not very sophisticated. However, it is simple to implement, and has reasonable chance of being trained with limited amount of data. With this project you are more interested in optimising a human's search time than it getting the best possible accuracy. Any accuracy better that random chance should be helpful.

Answer 2

The basic ways to make a recommendation engine are :

• Link a profile to similar ones
• Extract characteristics of a profile, and match them with characteristics of, here, jobs
• A mix of the two solutions above

Either way, you need data, labels, characteristics extractors...