Why You See So Much S*** On Your Twitter Timeline
Twitter open sourced their algorithm, and I have been peering under the hood for some fun.
One of the main questions that I was seeking some answers to was:
Why is my ‘For You’ timeline so full of s***?
This is an early chronicle for some of the potential reasons for it.
Too Long ; Didn’t Read takeaways
- If you are a passive User, Who You Follow and What They Interact With will have a heavy influence on your Timeline.
- If your Follows interact with polarized / incendiary / s****y topics, your timeline is likely to be filled with a lot of s***.
- If you are a passive User, and wish to clean up your timeline, one approach is to unfollow everyone, and tag individuals under specific Lists.
- Social media is tending towards algorithmically generated timelines. The main chokepoints in the algorithm’s value chain will become crucial in determining outcomes: for both influencers, consumers, and businesses that rely on social media for their livelihood.
Disclaimer: The code base is vast and has several interesting, amplifying implications. Errors, misinterpretations on this specific topic are entirely mine and are subject to revisions upon seeing the Light.
I. Core Problem: How Is Your ‘For You’ Timeline Built?
One of the main problem statements in algorithmically generated timelines is: How do you construct a timeline and prioritize what to show to the user?
Twitter tackles this in a three-stage process:
- Candidate Sourcing: Fetch the ‘best’ tweets from different sources.
- Ranking: Rank tweets using a Machine Learning model.
- Heuristic Filtering: Use a filter to tune out Blocked/NSFW content.
Out-of-Network
The first step — Candidate Sourcing — is crucial and a bedrock of everything that shows up in your For You timeline. This step is broken further into 3 sub-steps:
Sub-step 1: Twitter extracts the ‘best 1500 tweets’ from a pool of hundreds of millions of tweets.
Sub-step 2: In-Network sourcing: This draws ‘best tweet’ candidates from people you follow.
Sub-step 3: Out-of-Network sourcing: This draws ‘best tweet’ candidates from outside your network.
Your For You timeline today consists of 50% In-Network and 50% Out-of-Network sourced tweets.
As we think through this multi-layered sourcing challenge, the Out-of-Network sourcing is the most intriguing layer in the process. How can Twitter predict what tweets are relevant for you, if you do not follow them?
Twitter’s first approach to solve this problem is by stepping deeper into your In-Network, and attempts to gauge the nature of engagements of your Follows.
Twitter tries to find answers to these specific questions:
- What Tweets did the people I follow recently engage with?
- Who likes similar Tweets to me, and what else have they recently liked?
The second approach is based on Content Similarity: What Tweets and Users are similar to my interests?
If you are a passive User, who primarily lurks around Twitter for reading, the above approaches offer some clues to why a lot of s**t can show up in your timeline.
If you are a totally passive User who doesn’t engage with anyone, then Who You Follow and What They Interact With will start influencing what you see in your Timeline. Suppose you follow power users who happen to engage frequently with polarized / incendiary / Alice-in-wonderland topics, your Timeline will start resembling theirs over time.
This is one reason why you may feel your Timeline is filled with Negativity and generally s****y stuff.
II. Time To Clean Up Your Follows?
At the crux of Twitter’s Ranking algorithm is the following:
From Twitter:
This ranking mechanism takes into account thousands of features and outputs ten labels to give each Tweet a score, where each label represents the probability of an engagement. We rank the Tweets from these scores.
I have shown one possible representation of what the Ranking algorithm is likely to see from your Out-of-Network sourcing. If your Follows mainly engage heavily with Polarized / Incendiary topics, the Ranking algorithm will see a high cumulative composite score.
One of the black-box nature of the Ranking algorithm is the Coefficient Weights attached to each of the ten labels that determine the ranking.
Since the labels are not equal-weighted, some labels have an inordinate impact on the final ranking. Should these weights be changed in the future, your Timeline could change drastically.
Have you experienced a sudden spike in the visibility of ‘crazy’ tweets? This is most likely a manifestation of the Out-of-Network sourcing and Ranking at work.
One way for you to protect against this is to actively Show Less Often/Mute/Block individuals/topics.
Here is what the Ranking algorithm will likely see; assuming you Show Less Often/Mute/Block individuals on specific topics:
Doing this actively should clean up your Timeline.
A more effective approach is to unfollow everyone on your list, and add specific individuals/topics under specific Lists.
III. Social Media and Algorithmically Generated Timelines
Twitter deserves a lot of credit for open sourcing their recommendation algorithm.
Social media is tending towards algorithmically generated timelines. The main chokepoints in the algorithm’s value chain will become crucial in determining outcomes: for both influencers, consumers, and businesses that rely on social media for their livelihood.
As we evolve further along in this trend, algorithm value chain design will become one of the most crucial topics of interest for stakeholders in society.
