• Personalization is key to match Etsy’s unique marketplace with the right buyer at the right time. • Etsy introduced ADPM, a reusable three‑component deep learning module that learns from one‑hour user action sequences. • ADPM personalizes CTR and PCCVR ranking models for Etsy Ads, boosting relevance of sponsored listings. • Session personalization shows dramatic improvement: a user’s recent interest in men’s leather jackets changes ad outcomes. • Most shoppers view fewer than ten listings per hour, following a power‑law distribution that complicates personalization. • The challenge is met by leveraging rich user actions-search queries, favorites, views, add‑to‑carts-to infer intent. • This approach turns sparse session data into actionable signals, enhancing ad relevance across the platform.

Article Summaries:

  • Etsy has unveiled the adSformer Diversifiable Personalization Module (ADPM), a deep‑learning system that tailors sponsored listings to users in real time. By encoding one‑hour sequences of user actions-search queries, favorites, views, add‑to‑cart events, and purchases-ADPM generates a dynamic user profile that feeds into Etsy’s click‑through‑rate (CTR) and post‑click conversion‑rate (PCCVR) ranking models. The approach addresses the challenge that most sessions involve fewer than ten listings by leveraging the rich contextual signals from the action stream. Early results show more relevant ads, such as showing a user who recently viewed men’s leather jackets a personalized “jacket” search result set.
  • Etsy has unveiled a new personalization framework, the adSformer Diversifiable Personalization Module (ADPM), designed to tailor sponsored listings to users’ immediate interests. The system captures short‑term (one‑hour) sequences of user actions-search queries, item views, favorites, add‑to‑carts, and purchases-and feeds them into a reusable deep‑learning module. By generating a dynamic user representation from these streamed signals, ADPM optimizes click‑through and post‑click conversion ranking models in real time. Early demonstrations show that personalized rankings can surface more relevant listings, such as showing specific jacket styles to users who have recently viewed men’s leather jackets, improving relevance within the limited number of items typically seen in a session.
  • Etsy has unveiled the adSformer Diversifiable Personalization Module (ADPM), a deep‑learning system that tailors sponsored listings in real time. By encoding one‑hour “sessions” of user actions-search queries, item views, favorites, add‑to‑carts, and purchases-ADPM builds a dynamic user profile that feeds into its click‑through‑rate (CTR) and post‑click conversion (PCCVR) ranking models. The approach leverages both the content of listings and the sequence of user interactions, allowing the platform to adjust ad relevance on the fly. Early results show that personalization can shift sponsored results to better match a shopper’s intent, such as showing specific jacket styles after a user’s recent search refinements.
  • Etsy has unveiled a new deep‑learning framework, the adSformer Diversifiable Personalization Module (ADPM), to improve the relevance of its sponsored listings. The system encodes one‑hour “sessions” of user activity-search queries, views, favorites, add‑to‑carts, and purchases-into a dynamic user representation. By feeding this short‑term sequence into its click‑through‑rate (CTR) and post‑click conversion‑rate (PCCVR) ranking models, Etsy can adjust ad rankings in near real‑time. The approach tackles the challenge that most shoppers view fewer than ten listings per session by leveraging rich contextual signals from the session’s content and order of actions. The result is more intent‑aligned ad results for buyers and higher relevance for sellers.
  • Etsy has unveiled a new deep‑learning framework, the adSformer Diversifiable Personalization Module (ADPM), to improve the relevance of sponsored listings. ADPM builds a dynamic user profile from one‑hour “session” sequences of actions-search queries, item views, favorites, add‑to‑carts, and purchases-and feeds this into the platform’s click‑through‑rate (CTR) and post‑click conversion (PCCVR) ranking models. By combining content similarity with the temporal order of user interactions, the system can adjust ad rankings in real time, addressing the power‑law distribution where most shoppers view fewer than ten listings per session. The goal is to deliver more intent‑aligned ads for both buyers and sellers.

Sources: