Personalized Restaurant Ranking with a Two-Tower Embedding Variant | Towards Data Science

How a lightweight two-tower model improved restaurant discovery when popularity ranking failed