Upstage Wins Best Paper Honorable Mention at WSDM 2023

Upstage won the Best Paper Honorable Mention Award at WSDM 2023, the world-renowned conference on web search and data mining. This award is a prestigious honor given to only the top 0.5% of all submitted papers.

The research team, together with the research team from its Hong Kong subsidiary, presented research on session-based recommendation using the Atten-Mixer network. In session-based recommendation, which predicts a user's next action based on their recent behavior, this research proposes a method that improves recommendation performance by modeling user intent across multiple stages.

Existing GNN (Graph Neural Networks) approaches had the drawback that model complexity increased exponentially when attempting to improve performance. Upstage's research team overcame this limitation by removing parts of the neural network and optimizing the search space, simplifying the model while efficiently improving performance.

The method also improves recommendation quality by considering both the characteristics of items and the relationships between them. For example, when a user searches for a wedding dress and then looks for furniture, the system identifies the user's intent of 'getting married and decorating a new home' to accurately recommend related products.

This technology has already been applied to real-world services for two years, improving business metrics, and is being used to provide customized recommendation models across various industries through Upstage AI Pack, a no-code/low-code solution.