Introduction
Before the debacle, Twitter is a social media platform where users
like us can perform text messaging or “tweet” online and is widely known
for new-breaking tweets. At its launch, it was considered as “the SMS of
the internet” and was widely popular along with other social media
platforms such as Facebook, Instagram, and SnapChat.
Advertising and data licensing are the main source of revenue for
Twitter. Advertising services account close to 90% of Twitter’s revenue
which was $4.5 billion as of Q4 2021. The company generates their
advertising revenue by selling promoted products such as tweets,
accounts, and trends. By using their proprietary recommendation
algorithm, Twitter recommends ads and trends based on user’s interests
and tweets.
Recommendation Algorithm
Thank to Twitter for open-sourcing their algorithm, we can dive
deeper into their model and see how it works. The algorithm pipeline is
based on the tweets posted by users and consists of three following
stages:
- Candidate sourcing where best tweets are captured from different
sources. Those sources are based on the user’s both in-network and
out-of-network. The algorithm would favor toward the in-network and use
REAL GRAPH, an important component that predicts the possibility of
engagement with two users. The SOCIAL GRAPH captures the user
interaction in a time frame and is from the out-of-network source and is
another important component for the algorithm to capture. From there, a
custom matrix factorization algorithm is used to fetch the most liked
tweet from communities and help tailor the user’s home feed.
- With a pool of relevant candidates, a ranking of each tweets will be
compiled through a ML model that optimize for positive engagement.
- Heuristics and filters are being applied to provide an unique user
experience. Many features such as tweet diversity, content balance,
reduced negative tweets are implemented based on user’s preference and
outputs from prior stages.
Scenario Analysis
- The targeted audience is users that use their platform including
celebrities and politicians.
- Like other social media platforms, their key goal is to provide an
unique experience to users with personalized feeds and trends.
- I would suggest a better optimization to the home feed to not
overwhelm the user with numerous news and trends. To do that, trends and
news should be enlarged on the home feed for easier viewing.
Recommendation for Improvement
Fake news have been a biggest issue in social media platforms and
Twitter is no exception. For their recommendation algorithm, there needs
to be an additional stages where feeds need to be verified and
vetted.
Conclusion
In conclusion, recommendation algorithms serve a good way to provide
a better user experience when used correctly.