Large Language Models (LLMs) pose a new paradigm of modeling and computation for information tasks. Recommendation systems are a critical application domain poised to benefit significantly from the sequence modeling capabilities and world knowledge inherent in these large models. In this paper, we introduce PLUM, a framework designed to adapt pre-trained LLMs for industry-scale recommendation tasks. PLUM consists of item tokenization using Semantic IDs, continued pre-training (CPT) on domain-specific data, and task-specific fine-tuning for recommendation objectives. For fine-tuning, we focus particularly on generative retrieval, where the model is directly trained to generate Semantic IDs of recommended items based on user context. We conduct comprehensive experiments on large-scale internal video recommendation datasets. Our results demonstrate that PLUM achieves substantial improvements for retrieval compared to a heavily-optimized production model built with large embedding tables. We also present a scaling study for the model's retrieval performance, our learnings about CPT, a few enhancements to Semantic IDs, along with an overview of the training and inference methods that enable launching this framework to billions of users in YouTube.

The last decade has witnessed many successes of deep learning-based models for industry-scale recommender systems. These models are typically trained offline in a batch manner. While being effective in capturing users’ past interactions with recommendation platforms, batch learning suffers from long model-update latency and is vulnerable to system biases, making it hard to adapt to distribution shift and explore new items or user interests. Although online learning-based approaches (e.g., multi-armed bandits) have demonstrated promising theoretical results in tackling these challenges, their practical real-time implementation in large-scale recommender systems remains limited. First, the scalability of online approaches in servicing a massive online traffic while ensuring timely updates of bandit parameters poses a significant challenge. Additionally, exploring uncertainty in recommender systems can easily result in unfavorable user experience, highlighting the need for devising intricate strategies that effectively balance the trade-off between exploitation and exploration. In this paper, we introduce Online Matching: a scalable closed-loop bandit system learning from users’ direct feedback on items in real time. We present a hybrid offline + online approach for constructing this system, accompanied by a comprehensive exposition of the end-to-end system architecture. We propose Diag-LinUCB – a novel extension of the LinUCB algorithm – to enable distributed updates of bandits parameter in a scalable and timely manner. We conduct live experiments in YouTube and show that Online Matching is able to enhance the capabilities of fresh content discovery and item exploration in the present platform.

This papers presents a weakly supervised method to simultaneously address object localization and recognition problems. Unlike prior work using exhaustive search methods such as sliding windows, we propose to learn category and image-specific visual words in image collections by extracting discriminating feature information via two different types of support vector machines: the standard L2-regularized L1-loss SVM, and the one with L1 regularization and L2 loss. The selected visual words are used to construct visual attention maps, which provide descriptive information for each object category. To preserve local spatial information, we further refine these maps by Gaussian smoothing and cross bilateral filtering, and thus both appearance and spatial information can be utilized for visual categorization applications. Our method is not limited to any specific type of image descriptors, or any particular codebook learning and feature encoding techniques. In this paper, we conduct preliminary experiments on a subset of the Caltech-256 dataset using bag-of-feature (BOF) models with SIFT descriptors. We show that the use of our visual attention maps improves the recognition performance, while the one selected by L1-regularized L2-loss SVMs exhibits the best recognition and localization results.

We propose a novel method to address object localization in a weakly supervised framework. Unlike prior work using exhaustive search methods such as sliding windows, we advocate the use of visual attention maps which are constructed by class-specific visual words. Based on dense SIFT descriptors, these visual words are selected by support vector machines and feature ranking techniques. Therefore, discriminative information is learned and embedded in these visual words. We further refine the constructed map by Gaussian smoothing and cross bilateral filtering to preserve local spatial information of the objects. Very promising localization results are reported on a subset of the Caltech-256 dataset, and our method is shown to improve the state-of-the-art recognition performance using the bag-of-feature (BOF) model.