Slashdot Mirror
Using Machine Learning To Find a Better Job
An anonymous reader writes Artificial intelligence is gaining popularity in startups, with efforts ranging from virtual assistants to deep-learning approaches to business. An MIT Media Lab spinout called Beansprock is using natural language processing and machine learning to scour job listings for good matches for engineers and developers. The software breaks down users' skills and maps them to its understanding of open positions at companies. The site focuses on Boston-area jobs for now, but could expand if it's successful.
This company is run by two white men. They'll have an 18% better chance if they were women.
Move to India and get an H1B
They've automated the work of the drones that spam my linkedin profile after matching keywords and NOT reading anything in the free form text?
I apologize for the lack of a signature.
Look up FlipDog, a spinout of the WhizBang startup in 2001.
Got obsorbed by Monster.com.
Had the same description.
You won't find a job. Ever. Unemployment is about to go up and it won't stop. Automation will kill 99% of all jobs in the near future while in the present outsourcing and "efficiency directives" are destroying any hope you will ever have to make a living. And since the costs of living are going up, and not down, poverty is about to become a terrible reality for you and millions more. Oh, and contrary to what you may have been told by some armchair freedom fighters on the web, there is not going to be any revolution. Ever.
Well the first hiring decision is surely to hire a HR manager prepared to read the CVs, even if its only a scan read.
Because if HR won't read the CVs then why did you hire him? Did you read *his* CV?
I don't know, and perhaps it isn't so.
But the more machines control our gates,
The less we humans can control our fates.
But into the future we go!
Time is what keeps everything from happening all at once.
The next step will be to reverse the process and fire the hiring department?
... here in Europe (Germany and Austria) those requirements usually look like a letter to Santa Claus...
AI can parse job descriptions all day, but since most postings are either pure gibberish or outright lies, I'm not optimistic about the results.
Crap in -> Crap out.
But seriously, don't the common job sites do this already? I'm pretty sure a Monster; GlassDoor or LinkedIn already have departments working on better matching algorithms ( using NLP, supervised or un-supervised machine learning).
So they created a scraper and I'm supposed to be impressed?
This assumes that the job requisition and the resume are reasonably good descriptions of what is actually desired, etc.
My experience is that this is not the case. When you write the job req, it has to meet a variety of requirements imposed by policy, such as wording tied to specific salary grade competencies, responsibilities, etc. Then there's the "X years of experience with Y" kind of statement; which falls apart on several bases: was that 20 years of experience with C, or was it 20 repetitions of 1 year's experience; and it uses years of experience as a proxy for "skill level", which is hard to describe in objective words.
I'll concede that one can do much better than the sort of keyword searching that seems to be popular now, but building a reasoning engine that can read resumes and interpret them would be quite challenging. Resume reading is a skill that one develops with years of interviewing candidates: you learn to read between the lines: did they leave that job because the job ended and the company was winding down, or they found something better, or they were thrown out.
In other words, they've replaced the keyword-matching recruitment agent with a script? Way to go!
OK, I'm off to read the article now. Or at least the summary.
Free, as in your money being freed from the confines of your account.
Yeah, the people who need help getting jobs most are hipsters in the world's top startup hotspots. Great work, guys.
"When information is power, privacy is freedom" - Jah-Wren Ryel
I wrote a machine learning program that did when dice.com came online around 1990. I still use it today and it searches jobs (or any posting) on craigslist, monster, dice.com, hotjobs. It looks for keyword pattern matches in the job title, job description, etc, to give a score to a job posting, then if the job scores past a threshold, it will send a cover letter and email to the job. some of the most advanced features is that it will auto-generate sentences in the cover letter and/or job description depending on what it says in the job posting.
Interesting that this comes from MIT Media Lab - where you can find job descriptions like the following:
2. UNDEFINED DISCIPLINE
The Media Lab is a cross-disciplinary research organization focusing on the invention of new media technologies that radically improve the ways people live, learn, work, and play.
We are seeking a new kind of early career faculty member, not defined by discipline, rather by his or her unique and iconoclastic experience, style, and points of view. You can be a designer, inventor, scientist, or scholar – any combination – as long as you make things that matter. Impact is key.
This means somebody with at least these three sets of characteristics:
1) Being deeply versed in a minimum of two fields, preferably not ones normally juxtaposed;
2) Being an orthogonal and counter-intuitive thinker, even a misfit within normal structures;
3) Having a fearless personality, boundless optimism, and desire to change the world.
Any disciplines apply as long as their confluence shows promise of solving big, difficult, and long-term problems. And, most importantly, candidates must explain why their work really can only be done at the Media Lab. We prefer candidates not be similar to our existing faculty. We welcome applicants who have never considered academic careers. If you fit into typical academia, this is probably not the job for you.
The position has no specific degree requirement. Instead, candidates must show evidence of engineering accomplishment, scientific achievement, design innovation or artistic accomplishment. We are looking for a strong mix of invention, discovery and expression.
Applications should consist of one URL—the web site can be designed in whatever manner best characterizes the candidate’s unique qualifications. Web site should include a CV or link to a CV.
I wanted to nominate David Icke :)
One of the biggest problems is finding someone in HR who is willing to actually collaborate with the appropriate (knowledgable) people to craft a good positing. If conventional "I" can't read postings and figure out what the real job is about, then AI has no prayer of sussing out good patterns. I have done NLP and do ML on a regular basis -- there's no magic there. If the results are magical then your AI is locking onto nonsense inputs. As the data piles become ever larger, it becomes more common that any comparison you want to make is significant -- this leads to tremendous temptations for result cherry-picking. It's quite possible that a NLP-ML system could lock onto the quality of the grammar posting. I suppose it reflects on the company if they have high or low quality office staff in HR to manage the mechanics of posting jobs; but does that mean the job will be a better fit?
and, just one step in the right direction. But, it does mean being locked into a single job site or agency.
For me, a bigger issue is getting all the job descriptions (via email) that sound similar...but are being represented by multiple, usually, offshore, agencies. There are minor differences in the job descriptions that make it hard to know for sure they represent the same position. And, they don't like to tell you who the potential employer is UNTIL you sell them your first born and drink demon blood. They ALL want the right to represent you. Please be careful who you bargain with.. If you go with another agency for the same position (maybe offering a better rate) after they have submitted your resume to the potential employer, you will either not get the position or be locked in with the first one you gave your resume to. Hiring companies don't want to deal with the hassle of sorting out who gets the commission. In those situations, you lose.
Unscrupulous agencies will blast all resumes in their database that might fit a position in the hopes of being the first one in the door. You can be screwed without even knowing it if they have obtained your resume and submitted without your knowledge. They hope the employer sees one they like and calls them. Then, they reach out to you. It's one reason why automation is now used to sift through CVs and job applications with many finding their way to the bitbucket without ever being seen by a human. They get hundreds (maybe thousands) of resumes submitted electronically. You need to be standout. Being unique is one way to make it past that first cut. If not unique, you lose.
I have come view recruiters with a critical mindset after having dealt with some of the ilk that's out there. As a result, I keep a list of recruiters and agencies, gathered over the years, with who I trust to represent me well. There ARE good, reputable recruiters and agencies out there. Yes, I they are in the business of making money. And, I am very happy to help them do that if they work for me and help me gain employment on terms I prefer. This strategy, hopefully, prevents my CV from being rejected without human action first.
... would employ this software in their HR departments to better match prospective employees who've sent their resumes into the company to job openings that would actually be appropriate. I can't remember how many times I've received an email from a company that has my resume on file about a job that would be a "great match" only to read on and discover that the Venn diagram of the position's "gotta have" requirements, previous experiences, and technologies and those listed on my resume don't intersect. At all. I'm sure those companies eventually find someone to fill those roles but they'd find their ideal candidate a lot faster if they were doing a better job with their keyword searches. (Serious... is it really that difficult to set up an ATS?) Maybe MIT's work will help solve this problem for those companies. Or HR departments could stop trying to insulate the hiring managers from the hiring process so that boneheaded notices like the ones I received aren't being sent out. And making the company look completely clueless.
CUR ALLOC 20195.....5804M
Does machine learning overcome the HR requirement to have five years of experience in a technology that just came out six months ago?
How much are a dozen deep-learning researchers worth? Apparently, more than $400 million to Google with their DeepMind purchase:
http://www.technologyreview.co...
No, it fucking well isn't. There are no intelligent agents of any kind yet. If we keep watering the term "intelligence" down so that it is applied to anyone's algorithm that does anything handy, we're not going to have anything left to call an actual intelligence, which, among other things, will be conscious and that's why it will be intelligent. You know, able to generalize, capable of induction, intuition, experience, creativity, conversation, speculation, blue-skying, interest in how your day went, all that stuff -- the very thing that, when you meet a new fellow employee, makes you go "man, that bastard is intelligent!. And that won't be just because the entity knows H1-B is the path to corporate profit, either.
The only valid uses of AI we have at this point in time are of the form (a) "There's no AI yet" and (b) in the term "AI research." That's it. Otherwise, bupkis.
They are the most worthless fucks in all creation, I wouldn't let them wipe my ass. Think I'm trolling? every HRdroid I ever met could be replaced by a well designed "Welcome to Corp" video and a outsourced help line. Their chief skills are keeping management terrorized about worker revolts, which they themselves keep on the boil. Never work anywhere with an employee of the month, or casual friday, HR is more evil than Microsoft, there I've said it.