Reproduce “Suggest, Complement, Inspire: Story of Two Tower Recommendations at Allegro.com” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Aleksandra Osowska-Kurczab, Klaudia Nazarko, Mateusz Marzec, Lidia Wojciechowska, Eliška Kremeňová
Allegro.com implemented a unified content-based recommendation system based on the Two Tower framework, demonstrating its ability to generalize similarity, complementary, and inspirational product suggestions. A/B tests over two years revealed statistically significant improvements in user engagement metrics like CTR (up to +4.15%) and CVR (up to +2.22%), as well as increased GMV per visit (up to +0.31%).
Key insights
- A single Two Tower architecture can effectively generalize and unify distinct recommendation problems (similarity, complementary, inspirational) as variations of similarity search.
- Content-based features within a DLRM encoder effectively mitigate cold-start issues and manage dynamic product catalogs at scale.
- Hierarchical Approximate Nearest Neighbour (ANN) indexing can be used to introduce controlled diversity for inspirational recommendations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.