Knowledge Distillation (KD) is a powerful approach for compressing a large model into a smaller, more efficient model, particularly beneficial for latency-sensitive applications like recommender systems. However, current KD research predominantly focuses on Computer Vision (CV) and NLP tasks, overlooking unique data characteristics and challenges inherent to recommender systems. This paper addresses these overlooked challenges, specifically: (1) mitigating data distribution shifts between teacher and student models, (2) efficiently identifying optimal teacher configurations within time and budgetary constraints, and (3) enabling computationally efficient and rapid sharing of teacher labels to support multiple students. We present a robust KD system developed and rigorously evaluated on multiple large-scale personalized video recommendation systems within Google. Our live experiment results demonstrate significant improvements in student model performance while ensuring consistent and reliable generation of high-quality teacher labels from a continuous data stream of data.

We present the Learned Ranking Function (LRF), a system that takes short-term user-item behavior predictions as input and outputs a slate of recommendations that directly optimizes for long-term user satisfaction. Most previous work is based on optimizing the hyperparameters of a heuristic function. We propose to model the problem directly as a slate optimization problem with the objective of maximizing long-term user satisfaction. We also develop a novel constraint optimization algorithm that stabilizes objective tradeoffs for multi-objective optimization. We evaluate our approach with live experiments and describe its deployment on YouTube.

Multi-task prediction models and value models are the de-facto standard ranking components in modern large-scale content recommendation systems. However, they are typically optimized to model users’ passive consumption behaviors, and rank content in a way to grow only consumption-centric values. In this talk, we discuss the key insight that it is possible to model sparse participatory content-generation actions as well and grow ecosystem value through a new ranking system. We made the following key technical contributions in this system: (1) introducing ranking for content generation based on a categorization of user participation actions of different sparsity, including proxy intent action or access point clicks. (2) improving sparse task prediction quality and stability by causal task relationship modeling, conditional loss modeling and ResNet based shared bottom network. (3) personalizing the value model to minimize conflicts between different values, through e.g. ranking inspiring content higher for users who actively generate content. (4) conducting systematic evaluation of proposed approach in a large short-form video UGC (User-Generated Content) platform.

In recent years, social media users spend significant amount of time on Short-Form Video (SFV) platforms. Its success in creating an immersive viewership experience is not only from the content, but also due to its unique UI innovation: instead of providing choices for users to click, SFV platforms actively recommend content to users to watch one at a time. In this paper, we highlight unique challenges rooted from such UI changes for SFV recommendation system design. Firstly, there is yet much unexplored for sources of system biases under the new UI, as there are no clicks nor the common click-based position biases. Additionally, when training multiple types of user activities, positive labels for activities like sharing and commenting can be much sparser and more skewed than traditional click-based recommendation systems, as the latter can filter non-click impressions when generating "post-click" activities.

The sparse Mixture-of-Experts (Sparse-MoE) framework efficiently scales up model capacity in various domains, such as natural language processing and vision. Sparse-MoEs select a subset of the "experts" (thus, only a portion of the overall network) for each input sample using a sparse, trainable gate. Existing sparse gates are prone to convergence and performance issues when training with first-order optimization methods. In this paper, we introduce two improvements to current MoE approaches. First, we propose a new sparse gate: COMET, which relies on a novel tree-based mechanism. COMET is differentiable, can exploit sparsity to speed up computation, and outperforms state-of-the-art gates. Second, due to the challenging combinatorial nature of sparse expert selection, first-order methods are typically prone to low-quality solutions. To deal with this challenge, we propose a novel, permutation-based local search method that can complement first-order methods in training any sparse gate, e.g., Hash routing, Top-k, DSelect-k, and COMET. We show that local search can help networks escape bad initializations or solutions. We performed large-scale experiments on various domains, including recommender systems, vision, and natural language processing. On standard vision and recommender systems benchmarks, COMET+ (COMET with local search) achieves up to 13% improvement in ROC AUC over popular gates, e.g., Hash routing and Top-k, and up to 9% over prior differentiable gates e.g., DSelect-k. When Top-k and Hash gates are combined with local search, we see up to 100X reduction in the budget needed for hyperparameter tuning. Moreover, for language modeling, our approach improves over the state-of-the-art MoEBERT model for distilling BERT on 5/7 GLUE benchmarks as well as SQuAD dataset.

Multi-task learning aims to solve multiple machine learning tasks at the same time, with good solutions being both generalizable and Pareto optimal. A multi-task deep learning model consists of a shared representation learned to capture task commonalities, and task-specific sub-networks capturing the specificities of each task. In this work, we offer insights on the under-explored trade-off between minimizing task training conflicts in multi-task learning and improving multi-task generalization, i.e. the generalization capability of the shared presentation across all tasks. The trade-off can be viewed as the tension between multi-objective optimization and shared representation learning: As a multi-objective optimization problem, sufficient parameterization is needed for mitigating task conflicts in a constrained solution space; However, from a representation learning perspective, over-parameterizing the task-specific sub-networks may give the model too many ”degrees of freedom” and impedes the generalizability of the shared representation.

It is often desirable to model multiple objectives in real-world web applications, such as user satisfaction and user engagement in recommender systems. Multi-task learning has become the standard approach for such applications recently.

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.

Recommender systems are one of the most pervasive applications of machine learning in industry, with many services using them to match users to products or information. As such it is important to ask: what are the possible fairness risks, how can we quantify them, and how should we address them? In this paper we offer a set of novel metrics for evaluating algorithmic fairness concerns in recommender systems. In particular we show how measuring fairness based on pairwise comparisons from randomized experiments provides a tractable means to reason about fairness in rankings from recommender systems. Building on this metric, we offer a new regularizer to encourage improving this metric during model training and thus improve fairness in the resulting rankings. We apply this pairwise regularization to a large-scale, production recommender system and show that we are able to significantly improve the system's pairwise fairness.

Many recommendation systems retrieve and score items from a very large corpus. A common recipe to handle data sparsity and power-law item distribution is to learn item representations from its content features. Apart from many content-aware systems based on matrix factorization, we consider a modeling framework using two-tower neural net, with one of the towers (item tower) encoding a wide variety of item content features. A general recipe of training such two-tower models is to optimize loss functions calculated from in-batch negatives, which are items sampled from a random mini-batch. However, in-batch loss is subject to sampling biases, potentially hurting model performance, particularly in the case of highly skewed distribution. In this paper, we present a novel algorithm for estimating item frequency from streaming data. Through theoretical analysis and simulation, we show that the proposed algorithm can work without requiring fixed item vocabulary, and is capable of producing unbiased estimation and being adaptive to item distribution change. We then apply the sampling-bias-corrected modeling approach to build a large scale neural retrieval system for YouTube recommendations. The system is deployed to retrieve personalized suggestions from a corpus with tens of millions of videos. We demonstrate the effectiveness of sampling-bias correction through offline experiments on two real-world datasets. We also conduct live A/B testings to show that the neural retrieval system leads to improved recommendation quality for YouTube.

In this paper, we introduce a large scale multi-objective ranking system for recommending what video to watch next on an industrial video sharing platform. The system faces many real-world challenges, including the presence of multiple competing ranking objectives, as well as implicit selection biases in user feedback. To tackle these challenges, we explored a variety of soft-parameter sharing techniques such as Multi-gate Mixture-of-Experts so as to efficiently optimize for multiple ranking objectives. Additionally, we mitigated the selection biases by adopting a Wide & Deep framework. We demonstrated that our proposed techniques can lead to substantial improvements on recommendation quality on one of the world's largest video sharing platforms.

Neural-based multi-task learning has been successfully used in many real-world large-scale applications such as recommendation systems. For example, in movie recommendations, beyond providing users movies which they tend to purchase and watch, the system might also optimize for users liking the movies afterwards. With multi-task learning, we aim to build a single model that learns these multiple goals and tasks simultaneously. However, the prediction quality of commonly used multi-task models is often sensitive to the relationships between tasks. It is therefore important to study the modeling tradeoffs between task-specific objectives and inter-task relationships. In this work, we propose a novel multi-task learning approach, Multi-gate Mixture-of-Experts (MMoE), which explicitly learns to model task relationships from data. We adapt the Mixture-of-Experts (MoE) structure to multi-task learning by sharing the expert submodels across all tasks, while also having a gating network trained to optimize each task. To validate our approach on data with different levels of task relatedness, we first apply it to a synthetic dataset where we control the task relatedness. We show that the proposed approach performs better than baseline methods when the tasks are less related. We also show that the MMoE structure results in an additional trainability benefit, depending on different levels of randomness in the training data and model initialization. Furthermore, we demonstrate the performance improvements by MMoE on real tasks including a binary classification benchmark, and a large-scale content recommendation system at Google.

Much work has been done recently on learning word embeddings from large corpora, which attempts to find the coordinates of words in a static and high dimensional semantic space. In reality, such corpora often span a sufficiently long time period, during which the meanings of many words may have changed. The co-evolution of word meanings may also result in a distortion of the semantic space, making these static embeddings unable to accurately represent the dynamics of semantics. In this paper, we present a novel computational method to capture such changes and to model the evolution of word semantics. Distinct from existing approaches that learn word embeddings independently from time periods and then align them, our method explicitly establishes the stable topological structure of word semantics and identifies the surprising changes in the semantic space over time through a principled statistical method. Empirical experiments on large-scale real-world corpora demonstrate the effectiveness of the proposed approach, which outperforms the state-of-the-art by a large margin.

No abstract available.

Many recommenders aim to provide relevant recommendations to users by building personal topic interest profiles and then using these profiles to find interesting contents for the user. In social media, recommender systems build user profiles by directly combining users' topic interest signals from a wide variety of consumption and publishing behaviors, such as social media posts they authored, commented on, +1'd or liked. Here we propose to separately model users' topical interests that come from these various behavioral signals in order to construct better user profiles. Intuitively, since publishing a post requires more effort, the topic interests coming from publishing signals should be more accurate of a user's central interest than, say, a simple gesture such as a +1. By separating a single user's interest profile into several behavioral profiles, we obtain better and cleaner topic interest signals, as well as enabling topic prediction for different types of behavior, such as topics that the user might +1 or comment on, but might never write a post on that topic.

Many previous techniques identify trending topics in social media, even topics that are not pre-defined. We present a technique to identify trending rumors, which we define as topics that include disputed factual claims. Putting aside any attempt to assess whether the rumors are true or false, it is valuable to identify trending rumors as early as possible. It is extremely difficult to accurately classify whether every individual post is or is not making a disputed factual claim. We are able to identify trending rumors by recasting the problem as finding entire clusters of posts whose topic is a disputed factual claim.

Hundreds of thousands of hashtags are generated every day on Twitter. Only a few will burst and become trending topics. In this article, we provide the definition of a bursting hashtag and conduct a systematic study of a series of challenging prediction problems that span the entire life cycles of bursting hashtags. Around the problem of “how to build a system to predict bursting hashtags,” we explore different types of features and present machine learning solutions. On real data sets from Twitter, experiments are conducted to evaluate the effectiveness of the proposed solutions and the contributions of features.

Hashtags have been widely used to annotate topics in tweets (short posts on Twitter.com). In this paper, we study the problems of real-time prediction of bursting hashtags. Will a hashtag burst in the near future? If it will, how early can we predict it, and how popular will it become? Based on empirical analysis of data collected from Twitter, we propose solutions to these challenging problems. The performance of different features and possible solutions are evaluated.

Conventional studies of online information seeking behavior usually focus on the use of search engines or question answering (Q&A) websites. Recently, the fast growth of online social platforms such as Twitter and Facebook has made it possible for people to utilize them for information seeking by asking questions to their friends or followers. We anticipate a better understanding of Web users' information needs by investigating research questions about these questions. How are they distinctive from daily tweeted conversations? How are they related to search queries? Can users' information needs on one platform predict those on the other?

A Time Series Clique (TSC) consists of multiple time series which are related to each other by natural relations. The natural relations that are found between the time series depend on the application domains. For example, a TSC can consist of time series which are trajectories in video that have spatial relations. In conventional time series retrieval, such natural relations between the time series are not considered. In this paper, we formalize the problem of similarity search over a TSC database. We develop a novel framework for efficient similarity search on TSC data. The framework addresses the following issues. First, it provides a compact representation for TSC data. Second, it uses a multidimensional relation vector to capture the natural relations between the multiple time series in a TSC. Lastly, the framework defines a novel similarity measure that uses the compact representation and the relation vector. We conduct an extensive performance study, using both real-life and synthetic data sets. From the performance study, we show that our proposed framework is both effective and efficient for TSC retrieval.

A Time Series Clique (TSC) consists of multiple time series. In each TSC, the time series hold some natural relations with each other. In conventional time series retrieval methods, such natural relations are often ignored. In this paper, we formalize the problem of similarity search over TSC databases and develop a novel framework for similarity search on TSC data, which considers both time series patterns and relations. We conduct an extensive performance study, and the results show the effectiveness and efficiency of the proposed method.

The emergence of social media as a crucial paradigm has posed new challenges to the research and industry communities, where media are designed to be disseminated through social interaction. Recent literature has noted the generality of multiple features in the social media environment, such as textual, visual and user information. However, most of the studies employ only a relatively simple mechanism to merge the features rather than fully exploit feature correlation for social media applications. In this paper, we propose a novel approach to fusing multiple features and their correlations for similarity evaluation. Specifically, we first build a Feature Interaction Graph (FIG) by taking features as nodes and the correlations between them as edges. Then, we employ a probabilistic model based on Markov Random Field to describe the graph for similarity measure between multimedia objects. Using that, we design an efficient retrieval algorithm for large social media data. Further, we integrate temporal information into the probabilistic model for social media recommendation. We evaluate our approach using a large real-life corpus collected from Flickr, and the experimental results indicate the superiority of our proposed method over state-of-the-art techniques.