Recommender Techniques

Addressing the lengthy tail downside and enhancing the advice expertise for customers on the Headspace App

The Alternative Between Two Advice Techniques

Following weeks of diligent work, you lastly have a well-deserved night free to bask in a pleasant restaurant expertise. Two pals have provided restaurant suggestions. One good friend tends to stay to the tried and true, not often venturing into uncharted territory. As soon as a joke lands, they may repeat it incessantly with slight variations. Whereas their style is reliable, their suggestions have by no means left you completely impressed. This primary good friend mirrors a recommender system bothered by the lengthy tail downside; they advocate widespread eating places which might be reliable and palatable to the bulk however not essentially tailor-made.

Conversely, there’s the good friend who continually embraces danger, blurting out concepts in a haphazard method. You’re at all times left questioning about their style preferences. Whereas they’ve instructed some exceptional eating spots, they’ve additionally thrown in a fair proportion of dreadful ones, all delivered with the identical degree of enthusiasm. The second good friend is sort of a recommender system that throws out random strategies, often resulting in nice surprises but in addition disappointments. Whose recommendation would you observe for the night?

In my article, “Past Accuracy: Embracing Serendipity and Novelty in Suggestions for Lengthy Time period Consumer Retention,” I mentioned the significance of going past mere accuracy metrics to deal with the lengthy tail downside and improve the advice expertise for customers. On this article, I’ll focus on a real-world implementation once I labored as a machine studying engineer at Headspace Well being. Whereas there’s a paper revealed beneath the Artistic Commons Attribution 4.0 Worldwide License that goes in-depth with the implementation, I’ll clarify the rationale behind every resolution made and share the sudden findings we encountered through the course of.