TwiddleLab / Artificial Intelligence / Learning From Data
How Your Feed Picks: Recommendations
What you will learn
- A recommendation system suggests what you'll like from what people like you liked: collaborative filtering, kNN with viewers as points.
- Similarity can be as simple as counting matching ratings.
- A feedback loop narrows a feed: you click what it shows, it shows more of the same. New signals widen it.
How the lesson goes
You watched two cooking videos. Now your whole feed is cooking. Who decided that?
Make a bet: tap the video you think the app plays next. Then Check.
Teach the app your taste.
Like or skip each of the six videos. Your answers become a row in the app's table.
Find your taste twin.
Drag your row onto other viewers and see how many ratings you share. Then Check with the best match. Their ratings fill your ? cells.
This is how most feeds pick.
Tap to reveal it.
A recommendation system guesses what you'll like next.
Collaborative filtering: find people whose ratings match yours, and suggest what they liked. It's kNN again, with viewers as points.
A feedback loop: you click what it shows, it shows more of what you click.
Predict a rating from neighbours.
Mia hasn't seen the Space video. Drag the marker to the stars you'd predict for her. Then Check.
Now play 100 clicks.
Press Play. Each time, you click one of the six videos in your feed. The feed then shows more of what you clicked.
Pop the bubble.
Your feed is all cooking. You have 5 moves: Watch, Not interested, or Search. Get 4 different topics into your feed.
Code: how similar are two viewers?
Ratings live in a dictionary of dictionaries (Dictionaries: Labels, Not Numbers). Fill the two blanks, Run, then Check.
Which actions widen your feed, and which narrow it?
Drag each action to a side, or tap it and then a side. Then Check.
Your twin against the crowd.
Three picks from your taste twin, three of today's most popular videos. Like or skip each one, and see who knew you better.
Previous lesson: Biased Data, Biased Machine · Next lesson: Project: Be the Data Scientist · Part of the course: Learning From Data
Lessons are interactive and need JavaScript. Turn it on to play them.
