I'm building a testing project to get an introduction to DS & ML. As a person part of the working force, sometimes finding a job is harder than it should be. I thought I could built a testing project to help workers find a job that best match their interests and their skills.

I could use a classifier, but I do not know if regression if the best way to approach this as a first pass. I was thinking of a GA system that could find some way to find way to approach this over time in a X generation. Maybe this isn't the best way for recommending jobs.

What I'm looking for isn't the code for the problem but more of algorithms ideas I should take a look at and implement in any given programming language. I'm looking to implement a system that can take a peak at job descriptions, interests and skills of an individual. I think this doesn't sound too crazy to be able to train an agent and then make it look at every jobs there is on LinkedIn for instance and take a look at 'Software Engineer' and give me 45% matching and here's the three skills that you have to work on or 98% match, please apply.

  • 1
    $\begingroup$ Classification and regression are complements; one is not an algorithm to do the other. Try building a classifier based on the skills and recent job titles. Welcome to the site and good luck. $\endgroup$ – Emre Feb 22 '18 at 17:59
  • $\begingroup$ @Emre which classification algorithm would you recommend for this ? Bayes? $\endgroup$ – Kevin Avignon Feb 22 '18 at 18:13
  • $\begingroup$ Try a few so you can learn. Naive Bayes, logistic regression, neural networks... Once you have the features, it is not difficult to swap out the algorithm. $\endgroup$ – Emre Feb 22 '18 at 18:17
  • 1
    $\begingroup$ @Emre Seems slightly irresponsible to be spouting off models at this stage in the study design; I don't think your comments are providing a sound methodology. Please see my answer below. $\endgroup$ – I_Play_With_Data Feb 22 '18 at 22:46

You are much, much too early in your process to even begin thinking about your models. At this stage, you should be thinking about what the data looks like and how much of it do you need. For starters, what are you measuring? What is the answer that you seek? Is it job satisfaction? Likelihood to land an interview? Likelihood to get a job offer?

Once you have that, how are you quantifying that? Is it a categorical variable? A continuous variable? Then you would have to decide on the likely list of factors that go into that answer and determine how you're going to get that data. Do you need to run a custom survey? Are there other datasets that you can leverage? You need to think about all of these answers before you're ready for any sort of modeling discussion.

| improve this answer | |
  • $\begingroup$ I'm more interested about landing a job, but potentially a model that would be able to precisely say if I can land at least an interview is more appropriate since I can't know the heuristics a company uses to hire someone. What would you recommend to quantify this kind of problem? Thanks for all the tips concerning how to construct my training data set! $\endgroup$ – Kevin Avignon Feb 23 '18 at 13:01
  • $\begingroup$ @KevinAvignon well, which one is it? You can't apply the scientific method in a sound way and decide that you have two dependent variables (at least not in the same study). So you need to make some solid decision on what exactly it is you're pursuing, the scale on which that is measured and then go from there. You've posted a very nebulous idea, which is a start, but you need to provide more decisions & details for us to help you further. $\endgroup$ – I_Play_With_Data Feb 23 '18 at 14:35
  • $\begingroup$ The one I intend to pursue is the one that can model the use case which jobs I could land an interview at. I have no idea of the scale because I'm not really sure, but for now I think I'd keep it working only for Software Engineering jobs. What other stuff should I look into in my research ? $\endgroup$ – Kevin Avignon Feb 23 '18 at 15:00
  • 1
    $\begingroup$ @KevinAvignon OK so you can treat that as a Binary classifier of Interview/No Interview. Now you need to decide what each row of your data looks like. You already have your labels, so what goes into making those rows? What do you need to know about the applicant? What do you need to know about the job? Does this dataset exist or do you need to create it? My guess is that you're going to have a big mix of both continuous and categorical variables and will need to model appropriately once the dataset becomes clearer. $\endgroup$ – I_Play_With_Data Feb 24 '18 at 15:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.