Explore the concept of latent user intent modeling in recommender systems through this 25-minute conference talk from the Toronto Machine Learning Series (TMLS). Presented by Bo Chang, a Software Engineer at Google Brain, the talk delves into the challenges faced by current sequential recommender systems, which primarily rely on users' item-level interaction history to capture topical interests. Learn how these systems often lack a high-level understanding of user intent and why explicitly defining and enumerating all possible user intents is a complex task. Discover a proposed solution using latent variable models to capture user intents as latent variables through encoding and decoding user behavior signals. Gain insights into the practical application of this approach in a large industrial recommender system, and understand how it can potentially improve the accuracy and effectiveness of recommendation algorithms.