User-based vs item-based in ML Python - Model Approaches Compared

Model Pipeline - User-based vs item-based

This pipeline compares two popular recommendation methods: user-based and item-based collaborative filtering. Both use past user-item interactions to suggest new items, but user-based finds similar users, while item-based finds similar items.

Data Flow - 5 Stages

1Data Collection

1000 users x 500 items→Collect user ratings or interactions for items→1000 users x 500 items

User 123 rated Movie 45 as 4 stars

↓

2Preprocessing

1000 users x 500 items→Fill missing ratings with zeros or averages→1000 users x 500 items

Missing rating for User 10 on Item 20 replaced with 0

↓

3Similarity Calculation

1000 users x 500 items→Calculate similarity matrix (user-user or item-item)→1000 x 1000 (user-based) or 500 x 500 (item-based)

User 1 and User 2 similarity = 0.8 or Item 10 and Item 15 similarity = 0.75

↓

4Prediction

Similarity matrix and user-item matrix→Predict missing ratings using weighted average of neighbors→1000 users x 500 items with predicted ratings

Predicted rating for User 5 on Item 100 = 3.7

↓

5Recommendation

Predicted ratings matrix→Select top-N items with highest predicted ratings per user→1000 users x N recommended items

User 7 recommended items: [Item 23, Item 45, Item 78]

Training Trace - Epoch by Epoch

Loss
1.0 |*
0.9 | 
0.8 |*
0.7 | 
0.6 |*
0.5 |*
0.4 |*
0.3 |*
    +----------------
     1 2 3 4 5 Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	0.85	0.40	Initial predictions are rough with high error
2	0.65	0.55	Model learns better similarity weights, error decreases
3	0.50	0.68	Predictions improve, accuracy rises steadily
4	0.40	0.75	Model converges with good recommendation quality
5	0.35	0.78	Minor improvements, stable performance

Prediction Trace - 4 Layers

Layer 1: Input user-item ratings

Layer 2: Calculate similarity

Layer 3: Predict missing rating

Layer 4: Generate recommendations

Model Quiz - 3 Questions

What does user-based collaborative filtering use to recommend items?

A Similar users' preferences

B Similar items' features

C Random item selection

D User demographic data

Key Insight

User-based and item-based collaborative filtering both rely on similarity but differ in focus: user-based finds similar users to predict preferences, while item-based finds similar items to recommend. Both improve predictions by learning from neighbors, shown by decreasing loss and increasing accuracy.