Weekend Project: Build a Mini Recommendation System Using Snack Ratings (Ages 11–13)

Why this weekend project is perfect for ages 11–13

Recommendation systems power a lot of what kids see online: “You might also like…” on videos, songs, games, and even shopping. The cool part? Your child can build a tiny version at home with nothing more than snack ratings and simple math.

This is a beginner-friendly machine learning activity using ratings—no heavy coding required (though you can add it). It’s also a great conversation starter about how apps influence choices.

What your child will learn (without feeling like homework):

• What a recommender system is and why companies use it
• How data (ratings) becomes predictions
• The idea of similarity (“People who like what you like”)
• Basic ML habits: collecting data, testing, and improving

If you’ve been looking for fun AI experiments at home that feel practical, this one is a winner.

Materials + setup (20–30 minutes)

You can run this as a family activity with 2–5 people. More people = better recommendations.

You’ll need:

• 6–10 snacks (mix sweet/salty/healthy; include at least one “weird” option)
• Paper + pencil or a shared Google Sheet
• Optional: calculator

Snack ideas (aim for variety):

• Popcorn, pretzels, tortilla chips
• Chocolate squares, gummy candy
• Yogurt, fruit cup
• Granola bar, trail mix
• Something new (seaweed snacks, spicy chips, etc.)

Rules to keep it fair:

• Taste the same portion size (1–2 bites)
• Water between snacks
• No “influencing” other raters until after everyone rates

Pick a simple rating scale

Use a 1–5 rating scale:

• 1 = Nope
• 2 = Not my thing
• 3 = Okay
• 4 = Good
• 5 = Love it

Tip for parents: kids this age do great when you add one more choice: “Haven’t tried”. That prevents forced ratings.

Step 1: Collect snack ratings (your dataset)

This is the heart of the project. You’re building a tiny dataset like real recommendation systems do.

Create a table like this and fill it in as everyone tastes and rates.

Snack	Ava	Jayden	Mia	Parent	Notes (optional)
Popcorn	4	3	5	4	salty, crunchy
Chocolate	5	4	2	5	sweet
Seaweed	2	1	4	2	new flavor
Yogurt	3	5	3	4	creamy
Pretzels	4	2	4	3	crunchy
Gummies	5	4	3	4	fruity

Actionable tips to get better data:

• Include at least 6 snacks so the system has options.
• Try to avoid everyone rating everything the same (variety helps).
• If someone truly can’t try something (allergy), mark N/A and skip it in calculations.

Step 2: Build the “mini recommender” (two easy methods)

This is the “intro to recommender systems for kids” moment: turning ratings into suggestions.

You’ll build one of these approaches (or try both if your child is excited). Both are real ideas used in recommender systems.

Method A (easiest): “Similar Tastes” recommender

Goal: recommend snacks to a target person by finding the person whose taste is most similar.

How it works:

1. Pick a target person (example: Ava).
2. Compare Ava’s ratings to each other person.
3. Find the “closest match.”
4. Recommend snacks the closest match loved but Ava hasn’t tried (or rated low).

Simple similarity score (middle-school friendly): For each snack you both rated, compute:

• Difference = |Ava − Jayden|
• Total Difference = add all differences

The smaller the total difference, the more similar their tastes.

Example (Ava vs Jayden):

• Popcorn: |4−3|=1
• Chocolate: |5−4|=1
• Seaweed: |2−1|=1
• Yogurt: |3−5|=2
• Pretzels: |4−2|=2
• Gummies: |5−4|=1 Total Difference = 8

Do the same for Ava vs Mia, and Ava vs Parent. Lowest total wins.

Make a recommendation:

• Look at the most similar person’s top-rated snack(s) (4–5).
• If Ava rated those low (1–2) or hasn’t tried them, suggest them.

This method feels like: “Ava, you might like yogurt because Jayden has similar tastes and gave it a 5.”

Method B (still simple): Snack “leaderboard” recommender

Goal: recommend the best overall snacks based on average ratings.

How it works:

1. For each snack, add all ratings.
2. Divide by number of raters.
3. Sort from highest average to lowest.

This method is like: “Here are the top snacks in our family.”

Make it smarter (tiny ML upgrade):

• Instead of overall averages, calculate averages for a subgroup (e.g., “kids only” vs “parents only”).
• Compare differences: “Parents loved yogurt more than kids did—why?”

Which method should you choose?

• Choose Method A if your child likes detective work and “who matches who.”
• Choose Method B if your child likes clean numbers and ranking.

Both count as an AI project for middle school students because they model how real recommenders use data to predict preferences.

Step 3: Test your recommender (and improve it like real ML)

Now comes the fun part: see if your “AI” is actually helpful.

Run a prediction test

Pick 2 snacks your target person hasn’t tried yet (or rated a long time ago).

Testing steps:

• Your recommender predicts: “You’ll probably rate Snack X a 4 or 5.”
• The person tries Snack X.
• They rate it.
• Compare prediction vs reality.

Quick scoring:

• If prediction is within 1 point (predicted 4, actual 5), count it as a success.

Improve it (mini model upgrades)

Real recommendation systems are constantly tuned. Here are upgrades kids can actually do:

• Handle “N/A” correctly: only compare snacks both people rated.
• Add more data: add 3 more snacks or 1 more rater.
• Use “Top 3” recommendations: not just one snack.
• Add categories: label snacks as sweet/salty/spicy/healthy and see patterns.

A simple reflection checklist (great for ages 11–13)

Ask your child:

• Which person had tastes most similar to mine? Were you surprised?
• Which snack was hardest to predict?
• Did our ratings change after we talked about them?
• If this were a real app, what data would we not want it to collect?

That last question is sneaky-good: it introduces AI ethics without a lecture.

Next Steps: Make it bigger (and more “AI”) with Intellect Council-style extensions

If your child enjoyed this, you can turn it into a mini-series of weekend builds.

Option 1: Build it in a spreadsheet (fast, practical)

• Put ratings in Google Sheets.
• Add columns for differences and totals.
• Use AVERAGE() for Method B.
• Create a chart: snack averages as a bar graph.

Option 2: Try a beginner coding version (one step up)

• Use Scratch or Python (with help) to:
  • Store ratings in a table
  • Compute averages or similarity
  • Print: “Recommended snacks for Ava: …”

Option 3: Add a “confidence” score (smart and kid-friendly) Recommendations should be more confident when there’s more shared data.

• Confidence = number of snacks both people rated
• Example: “We’re 80% confident because you both rated 8 snacks.”

Option 4: Create a family “Recommender Challenge”

• Each person builds a recommender for someone else.
• Everyone makes 3 predictions.
• Track who gets the most accurate predictions.

How to get started today (15-minute plan):

• Pick 6 snacks and a 1–5 scale
• Make the ratings table
• Choose Method A or B
• Generate 2 recommendations
• Taste-test the recommendations and score accuracy

If your child is curious about what’s happening “under the hood,” this project is a perfect bridge into machine learning: real systems just use bigger datasets and more advanced ways to measure similarity. But the core idea—learning from ratings to make useful predictions—starts right here at the kitchen table.