Streaming media has become a popular medium for consumers of all ages, with people spending several hours a day streaming videos, games, music, audiobooks or podcasts across devices. Most global streaming services have introduced Generative Artificial Intelligence (GenAI) into their operations to personalize consumer experience, improve content, and further enhance the value proposition of streaming services. Despite the rapid growth, there is a need to bridge the gap between academic research and industry requirements and build connections between researchers and practitioners in the field. This workshop aims to provide a unique forum for practitioners and researchers interested in GenAI to get together, exchange ideas and get a pulse for the state of the art in research and burning issues in the industry.

With the exponential growth of video and other content across various domains including entertainment, e-commerce, education and social media, there is a growing need for personalized content recommendations that are relevant to users’ interests. However, building effective and scalable content recommender systems is challenging due to factors such as the vast volume of content, diversity of user preferences, inherent noise and bias in data, and the need for real-time recommendations.

At Peloton, we are challenged to not just surface relevant recommendations of fitness classes to our members, but also timely ones. As our fitness content library expands, we continually produce classes on certain themes which are most timely during a narrow time window. To address this challenge, we provide some control over our recommendations to external stakeholders, such as production and marketing teams. They enter timed boosts of certain classes during the windows they are relevant in. We have built out algorithms which take these desired classes and elevate the number of impressions for them, while preserving members’ engagement with our recommendations. In this paper, we discuss the system, the algorithms and some results from a few A/B tests showing how boosting works in practice.

Peloton is at the forefront of the at-home fitness market, with two business pillars: (a) a line of connected fitness equipment, and (b) a subscription-based service that offers access to a rich catalog of high quality fitness classes. As of May 2021, the total member base for Peloton stood at over 5.4 million, who took more than 170 million workouts. Peloton classes cover a diversity of instructors, languages, fitness disciplines, durations, intensity and muscle groups. On the other side, each user has their own specific fitness goals, time available to work out, fitness equipment and level of skill or strength. This diversity of content and individuality of user needs creates the need for a recommender system capable of personalizing the Peloton experience.

Recommending news and content is often more difficult than classic recommendation problems. At recommendation time, there is often less high quality explicit usage signals like upvotes, shares, dislikes, etc. because articles are relevant for a very short amount of time. Solely relying on implicit usage signals (views) in collaborative filtering for news articles often yields low quality documents optimized for views and clicks. Traditionally, content based filtering methods such as topic modeling, named entity extraction etc. are often used to counter or mitigate these issues but result in poorer recommendations on their own, and hybrid solutions of ensembles of content and collaborative filtering are difficult to optimize.

Dynamic topic models (DTMs) are very effective in discovering topics and capturing their evolution trends in time series data. To do posterior inference of DTMs, existing methods are all batch algorithms that scan the full dataset before each update of the model and make inexact variational approximations with mean-field assumptions. Due to a lack of a more scalable inference algorithm, despite the usefulness, DTMs have not captured large topic dynamics. This paper fills this research void, and presents a fast and parallelizable inference algorithm using Gibbs Sampling with Stochastic Gradient Langevin Dynamics that does not make any unwarranted assumptions. We also present a Metropolis-Hastings based $O(1)$ sampler for topic assignments for each word token. In a distributed environment, our algorithm requires very little communication between workers during sampling (almost embarrassingly parallel) and scales up to large-scale applications. We are able to learn the largest Dynamic Topic Model to our knowledge, and learned the dynamics of 1,000 topics from 2.6 million documents in less than half an hour, and our empirical results show that our algorithm is not only orders of magnitude faster than the baselines but also achieves lower perplexity.