Sequential models are invaluable for powering personalized recommendation systems. In the context of short-form video (SFV) feeds, where user behavior history is typically longer, systems must be able to understand users’ long-term interests. However, deploying large sequence models to extensive web-scale applications faces challenges due to high serving cost. To address this, we propose an industrial framework designed for efficiently serving large user sequence models. Specifically, the proposed infrastructure decouples serving of the user sequence model and the main recommendation model, with the user sequence model being served offline (asynchronously) with periodical refresh. The proposed infrastructure is also model-agnostic; thus, it can be used to support any type of user sequence models (even LLMs) with controllable costs. Empirical results show that large user models deployed with our framework significantly and consistently enhance the quality of the main recommendation model with minimal serving costs increase.

Recommender systems play an important role in many content platforms. While most recommendation research is dedicated to designing better models to improve user experience, we found that research on stabilizing the training for such models is severely under-explored. As recommendation models become larger and more sophisticated, they are more susceptible to training instability issues, i.e., loss divergence, which can make the model unusable, waste significant resources and block model developments. In this paper, we share our findings and best practices we learned for improving the training stability of a real-world multitask ranking model for YouTube recommendations. We show some properties of the model that lead to unstable training and conjecture on the causes. Furthermore, based on our observations of training dynamics near the point of training instability, we hypothesize why existing solutions would fail, and propose a new algorithm to mitigate the limitations of existing solutions. Our experiments on YouTube production dataset show the proposed algorithm can significantly improve training stability while not compromising convergence, comparing with several commonly used baseline methods.

Modern recommender systems have evolved rapidly along with deep learning models that are well-optimized for overall performance, especially those trained under Empirical Risk Minimization (ERM). However, a recommendation algorithm that focuses solely on the average performance may reinforce the exposure bias and exacerbate the “rich-get-richer” effect, leading to unfair user experience. In a simulation study, we demonstrate that such performance gap among various user groups is enlarged by an ERM-trained recommender in the long-term. To mitigate such amplification effects, we propose to optimize for the worst-case performance under the Distributionally Robust Optimization (DRO) framework, with the goal of improving long-term fairness for disadvantaged subgroups. In addition, we propose a simple-yet-effective streaming optimization improvement called Streaming-DRO (S-DRO), which effectively reduces loss variances for recommendation problems with sparse and long-tailed data distributions. Our results on two large-scale datasets suggest that (1) DRO is a flexible and effective technique for improving worst-case performance, and (2) Streaming-DRO outperforms vanilla DRO and other strong baselines by improving the worst-case and overall performance at the same time.

Recommender systems play an important role in modern information and e-commerce applications. While increasing research is dedicated to improving the relevance and diversity of the recommendations, the potential risks of state-of-the-art recommendation models are under-explored, that is, these models could be subject to attacks from malicious third parties, through injecting fake user interactions to achieve their purposes. This paper revisits the adversarially-learned injection attack problem, where the injected fake user ‘behaviors’ are learned locally by the attackers with their own model – one that is potentially different from the model under attack, but shares similar properties to allow attack transfer. We found that most existing works in literature suffer from two major limitations: (1) they do not solve the optimization problem precisely, making the attack less harmful than it could be, (2) they assume perfect knowledge for the attack, causing the lack of understanding for realistic attack capabilities. We demonstrate that the exact solution for generating fake users as an optimization problem could lead to a much larger impact. Our experiments on a real-world dataset reveal important properties of the attack, including attack transferability and its limitations. These findings can inspire useful defensive methods against this possible existing attack.

Many real-world recommender systems need to be highly scalable: matching millions of items with billions of users, with milliseconds latency. The scalability requirement has led to widely used two-stage recommender systems, consisting of efficient candidate generation model(s) in the first stage and a more powerful ranking model in the second stage.

Understanding temporal dynamics has proved to be highly valuable for accurate recommendation. Sequential recommenders have been successful in modeling the dynamics of users and items over time. However, while different model architectures excel at capturing various temporal ranges or dynamics, distinct application contexts require adapting to diverse behaviors.

We propose a novel way to train ranking models, such as recommender systems, that are both effective and efficient. Knowledge distillation (KD) was shown to be successful in image recognition to achieve both effectiveness and efficiency. We propose a KD technique for learning to rank problems, called ranking distillation (RD). Specifically, we train a smaller student model to learn to rank documents/items from both the training data and the supervision of a larger teacher model. The student model achieves a similar ranking performance to that of the large teacher model, but its smaller model size makes the online inference more efficient. RD is flexible because it is orthogonal to the choices of ranking models for the teacher and student. We address the challenges of RD for ranking problems. The experiments on public data sets and state-of-the-art recommendation models showed that RD achieves its design purposes: the student model learnt with RD has less than an half size of the teacher model while achieving a ranking performance similar tothe teacher model and much better than the student model learnt without RD.

User preferences are usually dynamic in real-world recommender systems, and a user»s historical behavior records may not be equally important when predicting his/her future interests. Existing recommendation algorithms -- including both shallow and deep approaches -- usually embed a user»s historical records into a single latent vector/representation, which may have lost the per item- or feature-level correlations between a user»s historical records and future interests. In this paper, we aim to express, store, and manipulate users» historical records in a more explicit, dynamic, and effective manner. To do so, we introduce the memory mechanism to recommender systems. Specifically, we design a memory-augmented neural network (MANN) integrated with the insights of collaborative filtering for recommendation. By leveraging the external memory matrix in MANN, we store and update users» historical records explicitly, which enhances the expressiveness of the model. We further adapt our framework to both item- and feature-level versions, and design the corresponding memory reading/writing operations according to the nature of personalized recommendation scenarios. Compared with state-of-the-art methods that consider users» sequential behavior for recommendation, e.g., sequential recommenders with recurrent neural networks (RNN) or Markov chains, our method achieves significantly and consistently better performance on four real-world datasets. Moreover, experimental analyses show that our method is able to extract the intuitive patterns of how users» future actions are affected by previous behaviors.

Top-N sequential recommendation models each user as a sequence of items interacted in the past and aims to predict top-N ranked items that a user will likely interact in a »near future». The order of interaction implies that sequential patterns play an important role where more recent items in a sequence have a larger impact on the next item. In this paper, we propose a Convolutional Sequence Embedding Recommendation Model »Caser» as a solution to address this requirement. The idea is to embed a sequence of recent items into an »image» in the time and latent spaces and learn sequential patterns as local features of the image using convolutional filters. This approach provides a unified and flexible network structure for capturing both general preferences and sequential patterns. The experiments on public data sets demonstrated that Caser consistently outperforms state-of-the-art sequential recommendation methods on a variety of common evaluation metrics.