A music-streaming app wants to help listeners find songs beyond their usual rotation. A recommender system can suggest music based on listening and interaction data, but its value depends on more than producing plausible suggestions: discovery should improve satisfaction, engagement, and ultimately retention. section 2.2.1 section 2.0
Recommendations personalize a product by predicting which items a user may want to explore. In an online shop, a suggestion might draw on previous purchases or other interactions. As people respond to suggestions, those responses can help refine what appears next. section 4.4
Choose an approach according to the data available and the experience you want to create:
- Collaborative filtering uses patterns among users or items. User-based filtering draws on similar users’ preferences; item-based filtering draws on items often viewed or purchased together. section 4.4.1
- Content-based filtering uses item attributes—such as style, brand, or price—to suggest products resembling those a user has engaged with. Early interactions can make it useful when a new user has little history. section 4.4.1
- A hybrid approach combines collaborative and content-based filtering rather than relying on one signal alone. section 4.5
From opportunity to experiment
A streaming team can assess whether years of listening, skip, like, and playlist-addition data make music recommendations feasible. A shopping team can then compare user-based and item-based strategies in an A/B test. Click-through rate indicates interest, while conversion rate tracks recommended items that become purchases; feedback and browsing behavior provide further signals for iteration. section 2.2.1 section 4.4.2
Monitor the experience after launch. What users want can depend on context, including season or where they browse. Test changes against user engagement and business outcomes rather than treating the initial model as finished. section 4.4.2
