Content Based Filtering

Building a recommender system that suggests items based on similar items’ attributes - the foundation of ‘items similar to this’ features.

Core Concept

This item is similar to other based on its attributes. If you liked this item, you might like these similar items.

1. Gather item attribute data (e.g., movie genres, product features)
2. Create item profiles based on attributes
3. Create user profiles based on their item preferences
4. Recommend items similar to those the user liked

Implementation Steps

• Gathered dummy item attribute data

data

• Created item profiles based on attributes

using tfidf

General Trivia: tfidf.fit_tranform(docs) -> leanrs vocabulary and IDF, returns document-term matrix (sparse matrix, shape(# of docs, # of unique words))

• Created user profiles based on their item preferences

User profile

General Trivia: linear_kernel() -> computes A.B(Transpose) (dot product). This equals to cosine similarity only if all the vectors are L2 normalized to unit length. cosine_similarity() -> explicitly computes cosine similarity and internally handles normalization.

• Recommended items similar to those the user liked using cosine similarity

recommend

Check my code: content-based-recommender.ipynb