Tiny Item-Item Recommender

A minimal, portfolio-friendly item–item collaborative filtering demo in Python. It builds a user–item matrix from a toy dataset, computes cosine similarity between items, and recommends top-N items for a chosen user.

How it works

1. User–item matrix: pivot ratings into a matrix (users as rows, items as columns). Missing ratings are treated as 0.

2. Item–item similarity: compute cosine similarity between item vectors:

   $$ \cos(\theta) = \frac{\mathbf{x}\cdot\mathbf{y}}{\lVert \mathbf{x}\rVert,\lVert \mathbf{y}\rVert} $$

3. Scoring: for each unseen item (j), [ \text{score}(j)=\sum_i \text{sim}(j,i)\times \text{rating}(u,i) ] then normalize by the total similarity weight to avoid popularity bias.

Quickstart

# 1) Create toy ratings (writes to data/ratings.csv)
python recommender.py --generate-data

# 2) Get top-5 recommendations for user U3
python recommender.py --user U3 --top_n 5

Use scikit-learn's implementation if you prefer:

pip install scikit-learn
python recommender.py --user U3 --top_n 5 --method sklearn

Example Output

=== Tiny Recommender ===
User: U3
 item_id  score
 I7       0.8123
 I2       0.7035
 I9       0.6461
 I4       0.5902
 I1       0.5718

(Number's will differ because toy data is different)

Files

recommender.py — all logic and a tiny CLI data/ratings.csv — toy ratings (auto-generated) requirements.txt — dependencies .gitignore — keeps the repo clean

Data format

Long-form CSV with three columns: user_id item_id rating U1 I3 4 U1 I7 5 U2 I1 2

Notes

Intentionally tiny and readable. Good for explaining item–item CF in interviews. Easy extensions: user-based CF, top-K neighbors, or a simple train/test split with evaluation metrics.