4 ms·

> Algorithmic feeds are always hostile to the user….They limit what a user sees…never encountered a situation where an algorithmic feed made things better I do

by dxbydt 5y ago

> Algorithmic feeds are always hostile to the user….They limit what a user sees…never encountered a situation where an algorithmic feed made things better I don’t know if Github should do algorithmic feeds. That said, my literal job description is to design an algorithmic feed for a tv news app. That’s what I’ve done - written an algo that decides what news gets shown, in what order => implicitly means a whole bunch of newsclips never get shown…or they are at the end of the queue and 90% of the viewers would never get to them. Why does one do this ? Because screen real estate is finite. You turn on your tv, you have a finite rectangular canvas - I can show maybe 5 squares on that rectangle without scrolling. Each square has a newsclip with text & picture. Who gets to decide which 5 clips to show ? Well, my algo does. How else would you do it ? There are literally 100s of 1000s of newsclips per day! Can’t show every single one of them - nobody has the time, nobody would watch, most of it is not going to be relevant to the vast majority of the audience. So 10000 newsclips get filtered down to 5 clips that get shown on the screen. You scroll. 5 more. Scroll again. 5 more. So on…you cap at 50 clips…nobody scrolls that much anyways. That’s why you have an algorithmic feed - because there is too much content, but your brain real estate is finite. Your screen real estate is finite. The amount of time you have to watch news is finite. Your interest in news is finite. All of this has been measured, analyzed, sliced & diced. Your time is finite. But content is infinite. There is just too much news. Take finance news - 100s of companies reporting earnings every given day, associated commentary around buying/selling that stock, how the company is doing…1000s of newsclips! Take political news - so much footage, all the way from local municipal town hall meetings to presidential press conferences. Take entertainment news…man, there is so much music/movie/art/talent related content. Can’t show everything. Hence algorithmic feed. Now sure, you can complain that the algo isn’t showing you what you personally want - its trained on the entire viewer population, not on your personal tastes. So most algos let you personalize the feed. But without an algo…there’s just no hope.

chockchocschoir 5y ago
> Now sure, you can complain that the algo isn’t showing you what you personally want - its trained on the entire viewer population, not on your personal tastes. So most algos let you personalize the feed. But without an algo…there’s just no hope. I think that the common complain is not "it isn't showing what I personally want" VS "it isn't showing what people generally want" but rather that the algorithm is often not tuned to "what X want" at all, but tuned to "what brings company most money". For example, YouTube recommendations are biased towards channels that have monetization activated, as YouTube earns more money that way, while ignoring channels the viewer (at large, or personally) would enjoy more.
- amelius 5y ago
  It's also "showing me stuff I may want when I least want to be aware of it". Stop the distractions already.
- edgyquant 5y ago
  If anything algorithms that drive engagement are mutually exclusive with what people want to see. Most people don’t usually engage with anything but the odds are a lot higher for someone to reply etc out of anger or disgust than anything else.
- dxbydt 5y ago
  Most of the comments here are incredibly dystopian - automation is bad, rehire the editor you've replaced at 75% of the salary, stop pushing engagement, what brings the company most money, megacorporation pushing slushy mass.... Jesus!! Man, didn't know recsys invokes such vitriol. There are dozens of news apps on smart TVs these days. Under the covers, it's pretty much the same deal - in go the raw feeds. Editors always do a first pass. Recsys algos do the 2nd pass. Ranking algos come in next. Personalization is usually last. Out comes the news. Viewers then have their say with thumbs up & downvotes. Those get fed back into personalization. On & on it goes... Its not magic. Its not perfect. But its how pretty much every smart tv news app works under the hood. There's dozens of them. Pick any one of them that works well on your TV for your temperament, I guess. Recsys is just a standard comp science discipline. We have our textbooks, our journals, our papers. We all read the same stuff, pretty much roll the same code with minor frills. Nobody's out to get you.
  - stormbrew 5y ago
    I don't have to think someone is "out to get me" to believe that the current state of affairs is deeply broken and that you can't fix it with a better mouse trap. This is just a really easy way for you to dismiss any criticism and avoid cognitive dissonance over your work. I believe you have good intentions. I also believe that those "dozens of news apps" all burn a whole bunch of watt hours on producing nearly identical results and are largely just facades over a handful of media companies. And that's not speculation or a conspiracy theory, it's plainly true. Media conglomeration has happened at an incredibly fast pace over the last 40 years and local media has been decimated. No one has to be a cackling evildoer with a fancy mustache and a british accent to make this happen, they just have to keep grinding harder and harder at inventing new ways to 'fix it' with more conglomeration.
newswasboring 5y ago
> what you personally want - its trained on the entire viewer population, not on your personal tastes. So most algos let you personalize the feed. But without an algo…there’s just no hope. This is a problem, but from a different point of view. The algorithm is trained not to give me news but to keep me engaged. That should not be the purpose of a news app.
jcelerier 5y ago
> So most algos let you personalize the feed. But without an algo…there’s just no hope. but there's an algo which is simple to implement and has very little biases: just show the things in temporal order ? Anything else is inferior to me. I don't want to see what's best for others, I don't want to see what some algorithm thinks is best for me, I just want to see what's last, with zero additional "reflection" on this.
- dxbydt 5y ago
  > just show the things in temporal order ? Anything else is inferior to me. heh heh. In the last 5 minutes, since you typed in your comment, there’s been 100 news clips. Literally. So just show the top 5 after reverse sorting by time….newest to oldest. Temporal order. Now what would be the problem with that ? Almost always, your newsfeed will end up dominated by the latest weather clip where the guy says “and now the temperature is a sunny 82 degrees”…meanwhile 100s of people die in ukraine but you didn’t see that because it was like an hour ago, and 1 minute ago >>> 1 hour ago.
  - ailef 5y ago
    If you also allow the user to filter results with a search query then this shouldn't be a problem. If he wants to see Ukraine news he will search for that. If there are too many results, the query can always be refined by the user.
- runako 5y ago
  "Temporal order" is another way of saying "prioritize accounts that post most frequently, while suppressing less-frequent posters." That obviously rewards e.g. unnaturally breaking up content into multiple parts, content-light posting, and other consumer-hostile behaviors. I'm not sure why so many seem to think this is obviously optimal for consumers. "I'll unfollow the accounts that behave that way" doesn't work because a temporally-ordered system incentivizes all accounts to behave this way. An account that posts highly valuable content, but only posts once weekly, will not thrive in a temporal-only medium.
  - 5y ago
stormbrew 5y ago
Honestly, there's nothing inherently wrong with providing a limited and selective set of information to people. The problem is the tendency towards broader and broader aggregation. Once upon a time, people got their news from approximately 3 "kinds" of places: - newspapers, locally managed, locally printed, locally distributed - radio, somewhere between locally and nationally managed, recorded, broadly distributed - television, probably regionally or nationally managed, maybe locally recorded, also broadly distributed. And for these things they had between several and tens of choices to select in to their preferences, interests, and alignments. They got to self-select the 'algorithm' they viewed information through. There was obviously an algorithm here, and in spite of golden-age glasses it was a racist, sexist, deeply problematic algorithm, but it was at least chosen by people in your community. Now though? Everything gets sucked up into some megacorporation's orbit through a rolling katamari of companies. We have facebook and twitter and things that will eventually be bought by facebook or twitter and approximately 2-3 companies that own every newspaper, radio station, and television station on the planet. There's no choice here anymore. Your algorithm is chosen by someone else, and if by some chance an upstart comes along to break out of that monotony it's inevitable that they'll be sucked up into the katamari eventually for the sin of being successful at bucking the trend, and then they will be just like everything else within a year. So you can do your best to provide a fresh, reasonable algorithm that does some good and helps people find something. But in the end, no matter how hard you try, it's gonna end up just like everything else and be a slushy mass of stuff that makes people angry to get money out of them.
hedora 5y ago
> How else would you do it ? There are literally 100s of 1000s of newsclips per day! No offence, but honestly, that scale doesn't require automation, and probably suffers from it. I'd rehire the full time editor (singular?) that you replaced, probably at 50-75% your salary.
eloisius 5y ago
I’m not sure what kind of news app you work on, so maybe this doesn’t fit, but what this makes me think is, you’ve optimized for putting the “best” stuff in limited real estate, but why does there need to be news headlines algorithmically ranked and dumped in front of me in a place that I’m not looking for them (e.g. the “home” screen of my smart TV)? If it’s an app where you are consciously seeking news headlines, maybe there’s a good case for collapsing the raw deluge of news into a recommended front page, but I still feel like we’re losing something. Recommendations are never good, and always corrupted by what the business wants to push on me, but even if they were good, the hyper personalized world is just weird. I’ll never be able to ask someone, “did you see the paper today?” Instead I wonder if they’ve been pushed the same stuff that I have and if we live in the same universe.
ars 5y ago
> So most algos let you personalize the feed. This is where the failure is. Most algos I've experienced show me more of what I already know. Or they show me what I was interested in the past (and I'm done with that). I want an algo that does the opposite - show me only things completely different - see if maybe I like them. I'm bored, I don't want more of the same, and I want something new. And there is where every one I've tried fails.

by dxbydt 5y ago

chockchocschoir 5y ago

> Now sure, you can complain that the algo isn’t showing you what you personally want - its trained on the entire viewer population, not on your personal tastes. So most algos let you personalize the feed. But without an algo…there’s just no hope. I think that the common complain is not "it isn't showing what I personally want" VS "it isn't showing what people generally want" but rather that the algorithm is often not tuned to "what X want" at all, but tuned to "what brings company most money". For example, YouTube recommendations are biased towards channels that have monetization activated, as YouTube earns more money that way, while ignoring channels the viewer (at large, or personally) would enjoy more.

amelius 5y ago
It's also "showing me stuff I may want when I least want to be aware of it". Stop the distractions already.
edgyquant 5y ago
If anything algorithms that drive engagement are mutually exclusive with what people want to see. Most people don’t usually engage with anything but the odds are a lot higher for someone to reply etc out of anger or disgust than anything else.
dxbydt 5y ago
Most of the comments here are incredibly dystopian - automation is bad, rehire the editor you've replaced at 75% of the salary, stop pushing engagement, what brings the company most money, megacorporation pushing slushy mass.... Jesus!! Man, didn't know recsys invokes such vitriol. There are dozens of news apps on smart TVs these days. Under the covers, it's pretty much the same deal - in go the raw feeds. Editors always do a first pass. Recsys algos do the 2nd pass. Ranking algos come in next. Personalization is usually last. Out comes the news. Viewers then have their say with thumbs up & downvotes. Those get fed back into personalization. On & on it goes... Its not magic. Its not perfect. But its how pretty much every smart tv news app works under the hood. There's dozens of them. Pick any one of them that works well on your TV for your temperament, I guess. Recsys is just a standard comp science discipline. We have our textbooks, our journals, our papers. We all read the same stuff, pretty much roll the same code with minor frills. Nobody's out to get you.
- stormbrew 5y ago
  I don't have to think someone is "out to get me" to believe that the current state of affairs is deeply broken and that you can't fix it with a better mouse trap. This is just a really easy way for you to dismiss any criticism and avoid cognitive dissonance over your work. I believe you have good intentions. I also believe that those "dozens of news apps" all burn a whole bunch of watt hours on producing nearly identical results and are largely just facades over a handful of media companies. And that's not speculation or a conspiracy theory, it's plainly true. Media conglomeration has happened at an incredibly fast pace over the last 40 years and local media has been decimated. No one has to be a cackling evildoer with a fancy mustache and a british accent to make this happen, they just have to keep grinding harder and harder at inventing new ways to 'fix it' with more conglomeration.

newswasboring 5y ago

> what you personally want - its trained on the entire viewer population, not on your personal tastes. So most algos let you personalize the feed. But without an algo…there’s just no hope. This is a problem, but from a different point of view. The algorithm is trained not to give me news but to keep me engaged. That should not be the purpose of a news app.

jcelerier 5y ago

> So most algos let you personalize the feed. But without an algo…there’s just no hope. but there's an algo which is simple to implement and has very little biases: just show the things in temporal order ? Anything else is inferior to me. I don't want to see what's best for others, I don't want to see what some algorithm thinks is best for me, I just want to see what's last, with zero additional "reflection" on this.

dxbydt 5y ago
> just show the things in temporal order ? Anything else is inferior to me. heh heh. In the last 5 minutes, since you typed in your comment, there’s been 100 news clips. Literally. So just show the top 5 after reverse sorting by time….newest to oldest. Temporal order. Now what would be the problem with that ? Almost always, your newsfeed will end up dominated by the latest weather clip where the guy says “and now the temperature is a sunny 82 degrees”…meanwhile 100s of people die in ukraine but you didn’t see that because it was like an hour ago, and 1 minute ago >>> 1 hour ago.
- ailef 5y ago
  If you also allow the user to filter results with a search query then this shouldn't be a problem. If he wants to see Ukraine news he will search for that. If there are too many results, the query can always be refined by the user.
runako 5y ago
"Temporal order" is another way of saying "prioritize accounts that post most frequently, while suppressing less-frequent posters." That obviously rewards e.g. unnaturally breaking up content into multiple parts, content-light posting, and other consumer-hostile behaviors. I'm not sure why so many seem to think this is obviously optimal for consumers. "I'll unfollow the accounts that behave that way" doesn't work because a temporally-ordered system incentivizes all accounts to behave this way. An account that posts highly valuable content, but only posts once weekly, will not thrive in a temporal-only medium.
- 5y ago

stormbrew 5y ago

Honestly, there's nothing inherently wrong with providing a limited and selective set of information to people. The problem is the tendency towards broader and broader aggregation. Once upon a time, people got their news from approximately 3 "kinds" of places: - newspapers, locally managed, locally printed, locally distributed - radio, somewhere between locally and nationally managed, recorded, broadly distributed - television, probably regionally or nationally managed, maybe locally recorded, also broadly distributed. And for these things they had between several and tens of choices to select in to their preferences, interests, and alignments. They got to self-select the 'algorithm' they viewed information through. There was obviously an algorithm here, and in spite of golden-age glasses it was a racist, sexist, deeply problematic algorithm, but it was at least chosen by people in your community. Now though? Everything gets sucked up into some megacorporation's orbit through a rolling katamari of companies. We have facebook and twitter and things that will eventually be bought by facebook or twitter and approximately 2-3 companies that own every newspaper, radio station, and television station on the planet. There's no choice here anymore. Your algorithm is chosen by someone else, and if by some chance an upstart comes along to break out of that monotony it's inevitable that they'll be sucked up into the katamari eventually for the sin of being successful at bucking the trend, and then they will be just like everything else within a year. So you can do your best to provide a fresh, reasonable algorithm that does some good and helps people find something. But in the end, no matter how hard you try, it's gonna end up just like everything else and be a slushy mass of stuff that makes people angry to get money out of them.

hedora 5y ago

> How else would you do it ? There are literally 100s of 1000s of newsclips per day! No offence, but honestly, that scale doesn't require automation, and probably suffers from it. I'd rehire the full time editor (singular?) that you replaced, probably at 50-75% your salary.

eloisius 5y ago

I’m not sure what kind of news app you work on, so maybe this doesn’t fit, but what this makes me think is, you’ve optimized for putting the “best” stuff in limited real estate, but why does there need to be news headlines algorithmically ranked and dumped in front of me in a place that I’m not looking for them (e.g. the “home” screen of my smart TV)? If it’s an app where you are consciously seeking news headlines, maybe there’s a good case for collapsing the raw deluge of news into a recommended front page, but I still feel like we’re losing something. Recommendations are never good, and always corrupted by what the business wants to push on me, but even if they were good, the hyper personalized world is just weird. I’ll never be able to ask someone, “did you see the paper today?” Instead I wonder if they’ve been pushed the same stuff that I have and if we live in the same universe.

ars 5y ago

> So most algos let you personalize the feed. This is where the failure is. Most algos I've experienced show me more of what I already know. Or they show me what I was interested in the past (and I'm done with that). I want an algo that does the opposite - show me only things completely different - see if maybe I like them. I'm bored, I don't want more of the same, and I want something new. And there is where every one I've tried fails.