Should Twitter algorithmically curate the timeline?

Zeynep Tufekci on Medium.com says no.  It seems Twitter is considering (instituting?) a method that would ignore strict reverse chronology, and if a user hasn’t accessed his or her timeline in a while, the more popular tweeters would be given some kind of priority in the queue.

She considers how the tweets about the death of Osama bin Laden spread so effectively, and from the account of a user (Keith Urbahn) who did not have many followers:

I honestly doubt that there is an algorithm in the world that can reliably surface such unexpected content, so well. An algorithm can perhaps surface guaranteed content, but it cannot surface unexpected, diverse and sometimes weird content exactly because of how algorithms work: they know what they already know. Yet, there is a vast amount of judgement and knowledge that is in the heads of Twitter users that the algorithm will inevitably flatten as it works from the data it has: past user behavior and metrics. I have witnessed Twitter network’s ability to surface unexpected content again and again, from matters small to large.

I suspect the really big news will get out very quickly under just about any reasonable algorithm.  The broader question is what kind of model we should use to consider Twitter curation.  Believe it or not, I am led to the thought of Ronald Coase.

As a reader, I seek an algorithm which weeds out some repetition.   For instance I sometimes see a Vox.com article in my feed from three different sources — it would suffice to see it once, along with a color shading indicating that some other people in my feed were tweeting the same thing.  I also would like blocks on tweets about the Super Bowl, Academy Awards, and so on.

That said, from a Coasean perspective, the tweeters may wish to impose these messages on me nonetheless.  Allowing users to create their perfect filters would in equilibrium mean that those sources send fewer other tweets through the system.  Some might leave Twitter altogether.  They are producing a service for free, and the ability to impose the bundle on me and other readers is part of what they value.  And indeed I also send self-promoting tweets (a justifiable practice provided it is not abused), and that is for me one reason to be on Twitter.  In other words, a major goal is to keep tweeters interested in supplying content, not to give every reader a perfect experience, and those two variables often conflict.

At the margin, should Twitter institute queuing rules to encourage the tweeters with many readers or the tweeters with relatively few readers?  The answer is not obvious.  The major tweeters produce more social value through their greater number of followers, but they may be reaping such high returns from being on Twitter that they don’t need added encouragement at the margin.  One approach is to prioritize well-regarded tweets, regardless of the number of followers of the tweeter.

For myself, I believe the ideal algorithm is to prioritize tweets from those who are “like” me in the sense of following similar people.  Or perhaps using similar grammatical constructions, or having tweeted similar links in the past.

Within these rules there are further opportunities for Coasean bidding for attention, using the @ function and also direct messages.

A separate issue is whether Twitter may wish to remedy the “overfishing” of the common pool of our attention which occurs when too many people tweet at peak time, and not enough people tweet at off peak times.  I suspect the demand for immediate gratification is too high for there to be gains from reshuffling the supply of tweets across time.

Overall I don’t see why company-regulated customization has to be a negative.  Tufekci put her anti-curation piece on Medium, which itself seems to have algorithms of curation, which in this case (fortunately) led me to her argument, wrong though it may be.

Comments

Honestly, who uses Twitter except narcissistic tweens, celebrities, and social butterflies? TC is a celebrity economist, yes.

" The herd is evil-smelling but it provides warmth." (Robert Ardrey)

I do, but the public timeline of the group I follow is normally unusable due to quantity. There are those people I follow due to kindness (I tend to keep mutual follows of friends) and those I follow b/c I'm interested in what they might tweet.

Lists are the filters that make the timeline more useful (for me).

He considers??
To be corrected to She.....

Yes, I thought that too. Zeynep is clearly a female name.

Since the marginal cost of a tweet is near zero, the resource will be overutilized precisely by those with the lowest opportunity cost of their time. These are not the droids we're looking for.

Who would follow those people?

'Believe it or not, I am led to the thought of Ronald Coase.'

So droll.

Comment curation: it is time.

Prof. Cowen certainly agrees - 'For myself, I believe the ideal algorithm is to prioritize tweets from those who are “like” me in the sense of following similar people.'

An echo chamber is apparently what the chairman and general director of the Mercatus Center would find a useful direction for a mass communication technology to follow. Considering just how often the linking here works in terms of following that idea, it is unsurprising to see it as his expressed preference.

And don't worry - comment curation is already practiced here anyways.

Oh, I think the comments here could stand a little more pruning....

Of all the racism these comments display, the common troll is your trigger?

I like the reference to Ronald Coase in the post … imagining what a hypothetical bargain would look like among the various users of Twitter can be a useful exercise … Also, whatever Twitter decides to do, it should avoid making the same mistakes Facebook has made.

Even though I think twitter was a brilliant idea and has a lot of potential as a media broadcasting tool, the site is terrible in many ways. 80-90% of people's followers are fake/inactive, and as evidence of this out of 1000 followers maybe two or three will re-tweet or favorite your stuff. The level of engagement relative to the number of followers is extremely low and probably due to a combination of all the fake and inactive accounts and algos. 140 character limit often results in poorly punctuated, incoherent tweets.Sometimes people will make multiple tweets at once to express a full thought and get around the the 140 limit. The future of twitter is less about individuals broadcasting tweets and more about receiving updates from very large, established accounts like celebrities and news sources

Soon, we can forget about 30 second sound bites mocked in yesteryear. In recent years, the 30 second sound bite has been reduced to the 10 second sound bite, and in Canada mere combinations of words drummed incessantly (e.g., "just visiting") appear to be sufficient to win elections.

The world will be ruled 140 characters at a time before you know it :)

In the ideal world the user would control his filters and could write his own or use available free software filters.

Computer enhanced gossip or social routing.

I was about to say that anything other than strict time ordering would be perceived as some sort of advertising-related manipulation. But your suggestion would make sense. Give the users tools to craft their own algorithm. It doesn't have to be in C -- it could be a GUI-based set of controls to throttle the effects of time-order, commentor popularity, snark, relevance, etc. And for the users who find that too complex, there could be presets -- you click a button for Most Popular Throttle Settings and they all get set to those values. And there's lots of potential there for other presets, such as Justin Beiber's Throttle Settings.

Curating content like this is almost always a disaster. Look at what the facebook 'newsfeed' is like.

In the digital age, I think companies should get used to the idea that the mere fact of having 1 billion users doesn't give them rights to try to use that to extract so many billions in revenues. Marginal costs are very near to zero, and enormous profitability can be achieved on the back of comparatively low revenues. Perhaps as a trust builder, it will eventually become necessary for some competitor to enter the market which formally offers strong legal guarantees to prevent future abuse of a user base.

For the idea of deciding "popular feeds", I would rather be in control of determining who I think is "popular". I may prefer to read recent twits from people I know or from specific microcosms of the twittosphere rather than those with the largest followings (and therefore with whom I am more likely to have a more mutually disinterested or indifferent relationship).

A system of up voting and down voting tweets that was recorded for each recipient, along with network analysis, would allow Twitter to build a personalized algorithm for each recipient. This would help them lock in their first moving advantages and economies of scale (they have the data to make the best user experience and other entrants don't) and would also be very useful for targeting which advertisements to show to each user.

I like the timeline the way it is. For a curated view of content, there are other tools, such as knowledgefilter.io, which algorithmically curates the _content_ referenced in tweets and other social media sources. (Disclosure: I work on this.)

I made wynno (https://wynno.com) to help cut down the noise in your Twitter timeline. You can make rules to filter out tweets that contain certain hashtags, keywords, links, etc. You can also vote up/down on tweets to train a personalized algorithm which will filter out tweets for you. Check it out! I hope you find it useful.

Comments for this post are closed