Personalized Fashion Discovery via Session Trajectories and Soft Negatives

Abstract

Fashion e-commerce routes every shopper through the same fixed category tree and reads a dislike as a permanent verdict. Both assumptions fail during exploratory browsing, where shoppers have no clear target and a “No” often means not this, right now. This work-in-progress note proposes navigation built per user, per session: the shopper is modeled as a single point in the item-embedding space, pulled toward likes and pushed from dislikes, every influence decaying over the session. The decay turns negative feedback into a soft negative: a rejected direction fades and can return, unlike the clean negative of Rocchio-style relevance feedback. A prototype on the public H&M catalog (≈ 105K articles) implements the full loop. We outline a two-stage evaluation (a simulation with drifting preferences, then a within-subjects user study) addressing interaction efficiency, the reading of negative feedback, and perceived user experience.