The idea
Content based recommendation uses item attributes rather than crowd behavior. It builds a profile of what a user likes from item features, then suggests new items with similar features.
Building it
- Describe each item by features, such as genre tags, keywords, or a text vector.
- Build a user profile by combining the features of items they liked.
- Score candidates by similarity between their features and the profile.
Strengths
- It works for a new item the moment its features exist, avoiding item cold start.
- It can explain recommendations through shared attributes.
- It does not need many other users to function.
Weaknesses
- It tends to recommend more of the same, limiting discovery, an effect called filter bubble.
- It depends on good feature quality, which is costly to curate.
- It cannot suggest something genuinely surprising the user might still love.
In practice
Many systems blend content based and collaborative signals into a hybrid to get the strengths of both.
Key idea
Content based recommendation matches item features to a user profile, handling new items well but tending toward narrow, similar suggestions.