Content-based filtering in ML Python

Introduction

Content-based filtering helps recommend items by looking at what you like. It finds similar things based on item features.

When you want to suggest movies similar to ones a user has watched.

When recommending books based on the genres a user prefers.

When suggesting songs that match the style of a user's favorite tracks.

When an online store wants to show products similar to what a customer viewed.

When a news app wants to recommend articles related to what a reader has read.

Syntax

ML Python

1. Collect features of items (like genre, keywords).
2. Represent items as vectors using these features.
3. For a user, find items similar to those they liked by comparing vectors.
4. Recommend the most similar items.

Features can be simple tags or complex descriptions.

Similarity is often measured using cosine similarity or distance metrics.

Examples

We compare items by counting common keywords to find similarity.

ML Python

# Example: Item features as keywords
item1 = ['action', 'adventure']
item2 = ['action', 'thriller']
# Compare similarity based on shared keywords

Text descriptions are converted into number vectors to compare items.

ML Python

# Example: Using TF-IDF vectors for item descriptions
from sklearn.feature_extraction.text import TfidfVectorizer
items = ['fast car racing', 'slow romantic movie']
vectorizer = TfidfVectorizer()
vectors = vectorizer.fit_transform(items)

Sample Model

This program shows how to recommend items similar to one the user liked using content-based filtering. It uses text descriptions and cosine similarity.

ML Python

from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity

# Sample items with descriptions
items = [
    'action adventure hero',
    'romantic comedy love',
    'action thriller spy',
    'romantic drama relationship'
]

# User liked item index 0 (action adventure hero)
user_liked_index = 0

# Convert item descriptions to vectors
vectorizer = TfidfVectorizer()
vectors = vectorizer.fit_transform(items)

# Calculate similarity of all items to the liked item
similarities = cosine_similarity(vectors[user_liked_index], vectors).flatten()

# Get indices of items sorted by similarity (excluding the liked item itself)
recommended_indices = similarities.argsort()[::-1][1:3]

# Print recommended items
print('User liked:', items[user_liked_index])
print('Recommended items:')
for i in recommended_indices:
    print(f'- {items[i]} (similarity: {similarities[i]:.2f})')

Important Notes

Content-based filtering only needs data about items, not other users.

It works well when item features are clear and descriptive.

It may recommend items similar to what the user already knows, limiting diversity.

Summary

Content-based filtering recommends items by comparing their features.

It uses item descriptions or tags to find similar items.

This method personalizes recommendations based on what the user liked before.