Twitter, recommendation engine approach

Twitter, as any social media environment has a mechanism to decide what content recommends you.

Right now, if you are using Twitter and are under “For you” tab, the content shown is recommended by an engine.

In the original map I have from Twitter this capability is shown here:

Twitter’s map, detail about “For you” recommendation

The recommendation code will be opened sourced

This week, this announcement was done: the recommendation code will be opened sourced.

What does it mean that the code will be outsourced?

It means many things, but mainly:

  • It’s the recognition that they are not able to solve it internally, so as in any case they want to solve it, they decide to open the code: it’s more important for them to solve the issue for the user than “to be right”.
  • It gives the opportunity to people outside of the organization to analyze the code and understand how these recommendations are done: transparency.
  • It gives Twitter the opportunity to receive feedback and findings of people that are studying the code and find issues on it.
  • It gives the opportunity to people to demonstrate their skills in front of a great challenge.
  • It accelerates the speed of twitter, as now they can:
    • Focus on the main changes on the product and let the people outside to find issues or other things.
    • Enable Twitter to receive feedback about it, that should be filtered for sure. Feedback is a treasure and have it will enable them to understand what is important for the user.
    • Focus on other product features while they are improving this too: I assume this has been getting a lot of energy from Twitter employees in the last weeks.
    • It gives Twitters employees a clear lesson about behaviors: it’s important to have results, if we are not able to obtain, we look for solutions outside. This is probably one of the more important aspects of this move and have a long term benefit. Twitter team is going from the world of “We know” to the world of “We do not know” and it’s powerful.

About transparency

Transparency has many dangerous edges, and till the moment that is normalized, Twitter will have a bumpy road.

  1. Transparency will show up some embarrassing issues (as commented on the tweet). This exposes Twitter’s team, they are vulnerable. This vulnerability, in the long run it’s positive.
  2. Transparency will show up some controversial decisions of the code that will be challenged. The decisions after these challenges will improve or make it worse the trust on Twitter.

An attempt to alter the map

I added temporary to the map just the accelerator symbol. The desire of act on it

Prioritize replies

To read the replies of a tweet that is full of answers is a key element of Twitter. Right now the answers are organized by creation date: first in, first shown. This gives the opportunity to some users that are able to reply early to be shown in very popular accounts.

So this is over:

Let’s see how it works.

What is clear to me is that the hierarchical structure of the users is the direction.

Update on March 31st

As promised, the code has been release on Github. Below the main diagram shown in the main page.

It’s going to be interesting to see what people can do with this.

Leave a Comment