On relevance reordering of Twitter Timeline

In the beginning of December 2015, people detected changes in Twitter’s home timeline (as result of experiments the company was likely running) causing once again a slew of press articles and a vigorous debate.

Twitter courts dislike by reordering tweets on 'relevance'

Just like in the case of the news feed changes on Facebook — the critics were as passionate as they were wrong.

For a full disclosure — I am having some apprehension writing this post — in part because I feel I have been too close to the cooker on this one, albeit some while ago. After all the project of re-engineering the home timeline is why I joined Twitter.

The topic of introducing smarts into Twitter timeline was a subject of the most passionate and prolonged internal debate that had it’s roots even before I joined the company in early 2012. Arguably the introduction of #Discover was an attempt to try out some ideas.

By early 2013 those of us on Twitter’s Search and Relevance team were able to fully crystalize a number of insights with respect to the dynamics of the product (not in a small part through the work of one of the data scientists on the team):

1. The home timeline was the only product that mattered on Twitter from a funnel standpoint.
2. The distribution of engagement in the home timeline relative to the age and competency of users (as measured for example by the number/quality of people they follow) had a very peculiar pattern. I remember drawing it on the white board during one specifically important meeting.
3. That pattern could be summed up as follows: we were doing a reasonable job engagement-wise only in a small slice of the curve of the user base that had ether been continually putting in a lot of effort maintaining their follower lists or simply ended up in that zone temporarily.

One interesting tidbit of insight was that active users routinely missed nearly all of the tweets in their home timelines.

We always felt that proponents of either “tweets only from my followers” or “Facebook-style relevance sort” question were presenting a false choice. The goal was not to limit — but to massively enhance the degree of control users would have over their timeline experiences. We believed it could be done through smart software and clever design focused on user’s information need, straightforward direct and indirect feedback mechanisms and absolute simplicity. The entire idea was to actually help users build and maintain the right network (and to do it without relying on excessive growth hacks).

We also felt that the best way to solve the problem was to tackle it directly instead of continuing to spend time on side product add-ons. As a good friend of mine frequently points out — the best way to get good at rock climbing is to just do the rock climbing.

After one long dinner with the Search and Relevance leadership team we formed a very small team to re-think the home timeline under the umbrella of the project “Highlight”. We set out to address 5 main problems in a way that mostly preserved time ordering and the overall feel of the product:

1. Make sure there always were interesting tweets in the timeline when the quality of the follower list wasn’t there.
2. Create a way for the users to contextually follow new people and accounts right from the home timeline.
3. Show users interesting tweets from their networks we knew they missed.
4. Create a way for the user to focus on most important (to that user) tweets.
5. Create an intuitive way in the home timeline to zoom-in into deeper content experiences — such as various types of events.

In just under a month we had a fully working product within the core Twitter mobile app that we were testing internally and externally with regular Twitter users (the rest of the story is probably best left for another time and other people to tell).

What is most surprising to me now — with a benefit of hindsight is how much we got right then — that and mixed feelings of somewhat of a lost opportunity at the time when massively bold changes to the product could have been most impactful.

The good news is — we are seeing new and improved versions of these features landing as complete experiences for all Twitter users — including the instant timeline, tweets you have missed and hopefully others. I trust the current Twitter team will stay the course, find a good way to integrate #Moments into the home timeline and rapidly continue implementing changes to the product that will make Twitter magical for all users.

I would like to extend my most sincere gratitude to all tweeps — former and current I had a privilege to work with and to wish everyone Happy 2016.

I would also like to thank Pankaj Gupta for encouraging me to write this post as a follow up to his excellent note about Twitter Moments.