Ask me a question about something you think I'd be very interested in. (Inspired by this post, but exactly the opposite:)https://twitter.com/visakanv/status/1090992506597081088 …
-
-
Replying to @generativist
A lot of data science is about capturing messy social concepts. For instance, a company that hires out temps might be interested in people who are 'reliable'. How does a computational social scientist go about constructing a measure like that?
1 reply 1 retweet 10 likes -
Replying to @kareem_carr
Incorrectly, I'd imagine. 'Reliability' is a damn good example of a "messy social concept." My (idealized) procedure would include first interviewing whoever does the hiring and whoever tasks the temps to see if I can elicit their loss functions for it.
1 reply 0 retweets 4 likes -
Replying to @generativist @kareem_carr
After that doesn't work, I'd probably ask for exemplars. Then, when that doesn't work because they're almost surely socially biased, I'd prob recommend temp task duplication so that I could forge some distance function.
1 reply 0 retweets 3 likes -
Replying to @generativist @kareem_carr
Finally, I'll end up on a conference call begging them to figure out what they want before I search for it. And, if it actually resolves well, I'd probably also go out and find a teammate who is really into causal inference
1 reply 0 retweets 3 likes -
Replying to @generativist @kareem_carr
This wasn't really meant to be a snarky answer. My experience with consulting really impressed upon me just how much of quantitative work is trying to identify what the person hiring you wants and what they think they want, and how rarely they are the same thing.
1 reply 0 retweets 2 likes -
Replying to @generativist
A temp agency probably just cares about maximizing profits. Proximate to that goal might be increasing the probability that a customer would use the service and maximizing the amount they would be willing to pay.
2 replies 0 retweets 1 like -
Replying to @kareem_carr @generativist
Coming up with a measure like 'reliability' is probably more for the legal system (to show lack of descrimination), the employee (so they feel like it's a fair system) and the customer (so they have criteria for comparing applicants and accepting recommendations).
1 reply 0 retweets 1 like -
Replying to @kareem_carr
I think this is a good tentative decomposition -- and, look how many different competing interests there are! But, it also kinda hints at some possible extant passive data that would be useful in the context of preference elicitation for each.
1 reply 0 retweets 0 likes -
Replying to @generativist @kareem_carr
I'd also probably try to ask someone like
@KLdivergence about known problems in this particular area first :)1 reply 0 retweets 1 like
Oh, also, my guess is that reliability for all stakeholders is contingent upon trust. Recursively. Trust generates reliability which generates trust which generates reliability which generates trust...
-
-
Replying to @generativist @KLdivergence
Trust is another concept we need to compute, but how? Credit card score is a version of this, but it's simplistic.
1 reply 0 retweets 1 like -
Replying to @kareem_carr @KLdivergence
Yes. And, I think people (myself included) are prone to sloppily conflating trust and reliability. They're different (albeit non-orthogonal) things.
0 replies 0 retweets 1 like
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.