Abstract: In implementations of systems and procedures for item recommendation and visualization, a computing device receives an input digital image depicting an environment and identifies attributes of objects within the environment. The attributes of the objects are used to identify items from item catalog data that have similar attributes and are suitable for inclusion within the environment. The identified items are displayed by way of a recommendation. The attributes of the objects are further used for generation of synthesized images. The synthesized images depict objects from the input digital image virtually replaced by items from the item catalog data. The synthesized images thus support visualization of the items within the environment depicted by the input digital image.

Abstract: Methods, systems, and computer storage media for providing an interest-driven recommendation engine in an item listing system are described. The interest-driven recommendation engine operates through three key stages: user interest modeling, item exploration, and recommendation delivery. In user interest modeling, user activity logs are used to build a hierarchical interest graph, capturing both short-term and long-term interests. A teacher-student Large Language Model “LLM” paradigm generates and scales these graphs efficiently, refining them to prioritize passion-driven interests over utilitarian needs. During item exploration, ranked interests are transformed into query expansions through a process called interest-to-recall. These queries are converted into embeddings, enabling the interest-driven recommendation engine to retrieve relevant items from the inventory.