The business problem
UrbanBites wants to add personalized recommendations to its mobile app. Two natural approaches: content-based (recommend items similar to ones you liked, based on item attributes) and collaborative (recommend items that people like you liked). Both are useful in different situations. Build both, compare results, and recommend which to ship — knowing each has its blind spots.
Two recommenders, one case
This case is split across two notebooks. Notebook 5A builds the content-based recommender; Notebook 5B builds the collaborative one. The interesting work is in the comparison: when does each fail, and what does that tell you about your data?
Case kit
Part A — Content-Based Recommender
- Notebook 5A — Content-Based Recommender
- Content-Based Excel Workbook
- Menu CatalogItem attributes used as the content features.
Part B — Collaborative Recommender
Sample executive deck
- Executive Recommendation DeckHow to position the two recommender approaches to leadership without losing them in the math.
Topics you'll be applying
- Decision 4 — Clustering — collaborative filtering is essentially clustering on user-item matrices
- Decision 6 — Text Analytics — content-based recommenders use text features from item descriptions