Content-based filtering recommends items similar to what the user liked in the past.
How It Works
- Build item profiles from features such as genre, keywords, text, metadata, or attributes.
- Build a user profile from items the user interacted with.
- Recommend items whose features match the user profile.
Example: if a user likes rock songs from the 1970s, recommend other rock songs from that era.
Advantages
- no item cold start when item features are available
- personalized to the user’s own history
- more explainable than many collaborative methods
- does not require other users’ data
Disadvantages
- limited discovery
- feature engineering is important
- new users still need initial preferences
- can overspecialize and create a filter bubble
Use Cases
- news article recommendation
- job postings
- academic papers
- domains with rich item metadata
This connects to Recommendation System Challenges because content-based filtering helps with new-item cold start but can worsen overspecialization.