[HN Gopher] Reducing Time-to-Hire and Finding Niche Candidates v...
___________________________________________________________________
Reducing Time-to-Hire and Finding Niche Candidates via Text Mining
and NLP
Author : cl42
Score : 22 points
Date : 2021-01-14 18:58 UTC (4 hours ago)
(HTM) web link (joinphase.com)
(TXT) w3m dump (joinphase.com)
| legerdemain wrote:
| I'm not sure... Scratch that. I'm _extremely skeptical_ that
| either job descriptions or resumes can be scoured for subtle
| indications of fit between jobs and candidates. If your new job
| isn 't quite what you expected or your new employee isn't working
| out quite the way you wanted, it's probably not because you
| didn't scour resumes and job descriptions minutely enough.
|
| Resumes are a very impoverished indicator of a candidate's
| abilities. So you "have experience" with a technology. Did you
| make the original decision to use it? Or were you asked to use
| it? Or was it in place when you got there? Were you only a
| consumer of its benefits? Do you understand its internals? Have
| you had to extend it? A technical project may take months or
| years, but your resume will represent it as a single line of text
| at best. "Year NNNN -- Company Foo -- Role Bar -- Implemented X
| using technology Y." Until resumes are structured like
| interviews, interviews are essential for getting this level of
| detail.
|
| Job descriptions are a very impoverished indicator of what a job
| is actually like. "We require N years of experience with X." Is
| that because this is your generic job description boilerplate? Or
| because your entire tech stack is inextricably tied to X? Will
| the candidate join a team that has inherited a legacy project
| that uses X? How is that team's morale? How is their turnover? If
| the candidate hates being on that team, how easy will you make it
| to switch? Are you hiring the candidate specifically with the
| goal of throwing people at a messy problem? Is it your
| responsibility to make sure the candidate succeeds? Until job
| interviews spend as much time revealing the company to the
| candidate as they spend grilling the candidate, back channel
| communication is essential for getting a job you like.
|
| I can't believe that ML and NLP offer new insights in this
| exchange. If I'm not willing to tell you something, you won't get
| it from me using NLP either. I've reached a point where
| recruiters telling me about using "cutting edge AI" results in
| their email getting deleted immediately. The last such service I
| responded to (Celential.ai) spammed me for a long time, and every
| job they sent me was garbage. Garbage to such a degree that it
| was my first experience with getting on the phone with a hiring
| manager and one or both of us saying, "No, there's really no
| point in us having this conversation, thanks," and hanging up.
| cl42 wrote:
| I agree with you and we're not trying to infer new things about
| people that aren't directly in the resume.
|
| Our approach is to take a more (unfortunately?) practical
| approach to this -- most recruiters and hiring managers simply
| scan most resumes for keywords, and most applicant tracking
| systems do the same... So it's not about predicting something
| from the resume, so much as making the matching process more
| flexible, given that most people doing the initial screening
| aren't doing a good job.
|
| In other words: make a very imperfect process slightly less bad
| today, so good candidates don't get missed.
|
| And of course, you still need to do the full set of interviews
| afterwards! We're not saying this is any sort of silver bullet.
| andy99 wrote:
| This is a point worth emphasizing. So many proposed "AI" use
| cases insist on trying to make some kind of end-to-end
| prediction, rather than let ML do something it is well
| positioned to do (e.g.) search more effectively, and let a
| person make the judgement for the things that aren't so
| readily predicted.
|
| My analogy is a weather forecast. Regular forecasts give you
| the temperature and precipitation, and let you decide what to
| do based on this. If meteorologists instead black boxed the
| weather and just tried to tell people what kind of clothes to
| wear every day, I suspect people would start ignoring the
| forecast. But this is how so many ML use cases are set up.
|
| I'm happy to see an application that respects ML for what it
| can do and not try and shoehorn it into the "prediction
| machines" paradigm.
| cl42 wrote:
| I love that analogy so much! Thank you for that.
|
| We'll definitely clarify that in future posts as well. We
| did our best to provide an example in the post, but it's
| tough. We got into talent/recruiting because I'm a former
| data science manager (my last startup was an AI one as
| well) and it always bothers me how little intentionality
| and awareness there is around limitations of AI in AI-
| driven products. Doesn't help that marketing teams tend to
| just promote the idea that "AI will solve everything",
| either!
| cl42 wrote:
| Hi folks! OP here + author of the NLP algorithms in the post. If
| you're hiring and want us to help you with filtering like this,
| we're definitely keen to pilot this or provide support in some
| form.
| karagenit wrote:
| Cool stuff - reminds me of a semester project I did about a
| year ago using latent semantic analysis to find relationships
| between documents. Sounds similar to what you guys are doing,
| trying to pair job descriptions with the best fitting resume.
| I'd love to hear more about the technical details on your
| solution, though I suspect that's somewhat of a "trade secret"
| ;)
| cl42 wrote:
| A big part of what we've been doing focuses on word embedding
| models and classifying individual parts of the document. The
| word embeddings help us look for synonyms general ideas
| around what the recruiter is looking for, versus what we are
| finding. We also score individual sections to see if the
| sections themselves cover all the bases or if it's the resume
| as a whole.
|
| We've also explored building our own universe of words/terms
| to help with this, but that's a 2.0 sort of dream, at this
| point!
| fatnoah wrote:
| I find this interesting. As someone with over 20 YOE, I've
| always eschewed the "2-page" resume guidelines. My full resume
| ends up being 4-5 pages, even with rigorous attention to only a
| 1 sentence description of each role, avoiding duplication of
| facts, and keeping and bullet points in the form of "Did X,
| resulting in Y".
|
| I've always wondered if a longer format helps me more with
| automated resume scoring. Also, does LinkedIn come into play
| here? I feel like when I show up to an interview, there's a
| 50/50 chance they're holding my resume vs. a printout of my
| LinkedIn profile.
| cl42 wrote:
| Yesssss... That's part of the issue here. I feel like people
| optimize for keywords because the applicant tracking systems
| today optimize for keywords. By the time a hiring manager
| gets to a resume, you've likely been filtered several
| times...
|
| The challenge is those filters suck. Hence our approach to
| making it "conversational" or flee-flowing. Using word
| embedding and other clustering helps a lot.
|
| Knowing nothing else, I'd say keeping your resume to 2 pages
| is critical... And make your "about" section on LinkedIn
| really stand out with a few key accomplishments.
|
| If you'd like, shoot me your resume at hello@phaseai.com and
| we can give you specific feedback.
| fatnoah wrote:
| I appreciate the offer to help, but my general success rate
| is high enough that I don't want to mess with stuff. I do
| target my pursuit of opportunities, and I ALWAYS rewrite my
| resume so that it prioritizes the experiences and skills
| most relevant to each job. If I were to ask for anything,
| it'd be something to highlight those matches.
| cl42 wrote:
| Amazing. If I had a penny for every job seeker who told
| me they don't do any resume customization... I'd have
| lots of pennies! :)
| sqrt17 wrote:
| err ... it's kind of what you'd get from Textkernel or LinkedIn
| Recruiter, probably at a similar price. But with nicer fonts,
| so ok.
___________________________________________________________________
(page generated 2021-01-14 23:01 UTC)