Researchers at the Cornell and the Technische Univerität Berlin and Cornell have studied the problem that more popular items get priority in search results, creating a positive feedback loop that unfairly deprecates other, equally valuable items.
Rankings are the primary interface through which many online platforms match users to items (e.g. news, products, music, video). In these two-sided markets, not only the users draw utility from the rankings, but the rankings also determine the utility (e.g. exposure, revenue) for the item providers (e.g. publishers, sellers, artists, studios). It has already been noted that myopically optimizing utility to the users – as done by virtually all learning-to-rank algorithms – can be unfair to the item providers. We, therefore, present a learning-to-rank approach for explicitly enforcing merit-based fairness guarantees to groups of items (e.g. articles by the same publisher, tracks by the same artist). In particular, we propose a learning algorithm that ensures notions of amortized group fairness, while simultaneously learning the ranking function from implicit feedback data. The algorithm takes the form of a controller that integrates unbiased estimators for both fairness and utility, dynamically adapting both as more data becomes available. In addition to its rigorous theoretical foundation and convergence guarantees, we find empirically that the algorithm is highly practical and robust.
Journal Reference:
Marco Morik, Ashudeep Singh, Jessica Hong, and Thorsten Joachims. 2020. Controlling Fairness and Bias in Dynamic Learning-to-Rank. In Proceedings
of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR '20), July 25–30, 2020, Virtual Event, China.
ACM, NewYork, NY, USA. DOI: https://doi.org/10.1145/3397271.3401100
Maybe this, if deployed widely, can help reduce the tendencies for discourse to develop isolated silos.
(Score: 2) by FatPhil on Tuesday August 25 2020, @02:30AM (7 children)
The problem is that good-enough quickly-enough often-enough items are given priority (so make quicker) in search results. The optimisation is for adequacy, not for perfection.
However, it sounds like their solution doesn't even address either of these "problems". To me, this symmetric evaluation function looks like it's intended to deliver sub-optimat results to the human, because some websites would prefer to get those lovely lovely clicks. Which sounds just terrible.
Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
(Score: 2) by The Mighty Buzzard on Tuesday August 25 2020, @03:09PM (6 children)
Adequacy is by definition adequate though.
My rights don't end where your fear begins.
(Score: 2) by FatPhil on Tuesday August 25 2020, @07:29PM (5 children)
Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
(Score: 2) by The Mighty Buzzard on Wednesday August 26 2020, @02:16PM (4 children)
Generally? Yes. It's a step up from what I usually find.
If I'm on a search for the best burger in TN because I want to know where to get it, I expect to have to wade through a whole lot of garbage before I find it. If I'm simply looking for an acceptable burger for lunch so I can eat and get back to work, I do not want to have to sample twenty or thirty burgers before I find an acceptable one; I want adequate and I want it the first time. Which is exactly what I want out of search engines five nines of the time.
My rights don't end where your fear begins.
(Score: 2) by FatPhil on Wednesday August 26 2020, @02:39PM (3 children)
But it's shitty, as the things that were good for you were drowned in things that were not good.
Adequate is typically pretty shitty, that's what I'm trying to say, and definitely nothing to strive for.
Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
(Score: 2) by The Mighty Buzzard on Wednesday August 26 2020, @03:10PM (2 children)
See, I define adequate as adequate. Acceptable. Good enough. I don't define it as "typically shitty".
My rights don't end where your fear begins.
(Score: 2) by FatPhil on Thursday August 27 2020, @11:07AM (1 child)
Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
(Score: 2) by The Mighty Buzzard on Saturday August 29 2020, @05:22PM
Just did. "Adequate is typically pretty shitty" sums your position up quite nicely. And I disagree because I do not accept anything with a significant amount of shitty as adequate.
My rights don't end where your fear begins.