How does Twitter use machine learning? by Arindam Paul
Answer by Arindam Paul:
I do not work on Twitter so I cannot deterministically answer this. I hope someone from Twitter answer this. However, as a person researching in machine learning, I can try to guess. Although I think they do not use it as much as Facebook or Google does as there is not much moderation or ranking going on.
- Trending: So twitter has this thing at the right (at least on my chrome desktop browser). You can see that what trends does not depend on what has the highest tweets right. So, I presume they use some machine learning to automatically learn the time-series data (tweets)
2. Moments: This is where there is not much difference between Twitter and a news website like Times. I believe (again I do not work at Twitter) that there is some kind of ML and heuristic algorithm which determines what pops up as Moments. One parameter which is definitely used is location.
3. Who to Follow: Again this uses information about the things you post (textual information, hashtags) , people following you, people recenty followed by you, people whose posts you favorited.
Also, I know for sure lot of people use twitter data for machine learning. I guess most of you knew this already. I had worked on a course project where I looked at "I just bought" tweets when people buy things from Amazon.
One arena where they do not use machine learning is modulate people's feed. So Twitter does not organize the feeds based on who you recently followed, who responds to your tweets etc.