The Search-Recommendation-Notification Spectrum

The key themes that tie together search engines, recommender systems and smart push notifications

By Nikhil Dandekar, Engineering Manager, Search at Foursquare

There has been a lot of talk recently, including at the last RecSys about the convergence of search and recommendations. Having worked in this area for several years now, I feel search and recommendations are part of the same spectrum, two problems with the same overarching structure. What’s becoming more obvious to me over time, is that a large subset of smart push notifications (sent on your phone, or your smartwatch, or your wearable device) also lie along the same spectrum.

In all these cases, there is some smart content that the app wants to show the user. The user either explicitly asks for the content (search) or the app deduces that the user might want the content based on the context (recommendations and notifications).

For recommendations, the context is usually the page or screen the user is on (e.g. the Netflix home screen, or Twitter people to follow). For push notifications, the context is usually a specific trigger like being at a particular location at a particular time (Foursquare) or achieving your daily goals of steps (Google Fit) etc.

As we go from search to recommendations to notifications, the user intent becomes harder and harder to deduce. At one end are specific, navigational searches (e.g. searching for “youtube” in Google) which have a very clear intent and a single right answer. On the other end are push notifications, where the user never expresses any intent. Here the app guesses the user might have an intent and pushes content to the user.

There are a set of themes that are common for the range of problems along this spectrum. Understanding these themes is important if you want to build a good relevance product. The following list captures most of the big ones:

1. Delivery mode: How do you deliver content to the user?
2. UX: What’s the right UX for the product?
3. UX fault tolerance: How fault-tolerant does the UX need to be?
4. Personalization: Does personalizing the experience matter?
5. Result justifications: How well do you need to explain why each of the results are returned?
6. Result diversity: Do you need to showcase diverse categories of results?
7. Result freshness: Does the content need to change frequently for the product to be compelling?
8. Serendipity: What’s the possibility of being surprised or pleasantly delighted?
9. User tolerance: Will users be ok with low-quality content?

Here’s how each of these themes differ between search, recommendations and notifications.

Keep these themes in mind the next time you build one of these products.