The escalating complexity of industrial recommendation systems, characterized by diverse user behaviors and cross-domain application scenarios, necessitates advanced multi-task and multi-domain learning paradigms. Existing methods often struggle with efficient knowledge transfer across tasks and domains due to semantic gaps and distribution shifts. To address these challenges, we propose the Perception-Aware Multi-Task Transformer Network for Personalized Multi-Domain Adaptation (PMTA), a unified framework that integrates three key innovations: First, the Task Prompt Encoding (TPE) module dynamically generates prompts by synthesizing personalized user data with task-specific information. Second, the Transformer-based Multi-Task Perception (TMPN) network enables adaptive cross-task knowledge transfer through attention mechanisms. Third, the Multi-Domain Adaptation (MDAN) component captures domain-specific behavior patterns via learnable prior information. Experimental results demonstrate PMTA's effectiveness, achieving 0.168% increase in watch time and significant improvements in engagement metrics (AAD: +0.0113%, AAH: +0.0608%). Deployed on Douyin and Douyin Lite, it significantly improves recommendation quality and drives commercial success.

Live streaming has emerged as a dynamic content format that delivers real-time and interactive experiences to users. Distinguished by the short lifespan and immersive nature of live rooms, live streaming poses two key challenges for recommendation: (1) Timeliness: the model must rapidly identify and promote relevant live rooms to target users within a limited window; and (2) Accurate staytime prediction: since extended watching often reflects content quality and user satisfaction, precisely predicting staytime serves as a critical indicator of recommendation relevance and user engagement. Existing approaches often improve timeliness by repeatedly sending staytime signals to accelerate model learning. However, this introduces label truncation bias, distorting the unbiased estimation of high staytime samples. To reconcile these competing demands, we propose MS3M (Multi-Stream Segmented Staytime Modeling), a novel framework that leverages multiple data streams for faster learning while employing segmented staytime modeling-converting staytime regression into a series of time-segmented classification tasks to ensure unbiased training. Furthermore, to address the sparsity of high staytime samples, MS3M's task-dependent architecture allows high staytime parameters to leverage prior knowledge from low staytime data, significantly improving generalization for long-duration watching behaviors. Extensive offline experiments and online A/B tests on TikTok confirm that MS3M effectively balances timeliness and unbiased learning, leading to substantial gains in recommendation accuracy. The proposed approach currently serves TikTok's live streaming recommendation system, contributing to continuous improvement in user watching experience.

Recent progress on large language models (LLMs) has spurred interest in scaling up recommendation systems, yet two practical obstacles remain. First, training and serving cost on industrial Recommenders must respect strict latency bounds and high QPS demands. Second, most human-designed feature-crossing modules in ranking models were inherited from the CPU era and fail to exploit modern GPUs, resulting in low Model Flops Utilization (MFU) and poor scalability. We introduce RankMixer, a hardware-aware model design tailored towards a unified and scalable feature-interaction architecture. RankMixer retains the transformer's high parallelism while replacing quadratic self-attention with multi-head token mixing module for higher efficiency. Besides, RankMixer maintains both the modeling for distinct feature subspaces and cross-feature-space interactions with Per-token FFNs. We further extend it to one billion parameters with a Sparse-MoE variant for higher ROI. A dynamic routing strategy is adapted to address the inadequacy and imbalance of experts training. Experiments show RankMixer's superior scaling abilities on a trillion-scale production dataset. By replacing previously diverse handcrafted low-MFU modules with RankMixer, we boost the model MFU from 4.5% to 45%, and scale our online ranking model parameters by two orders of magnitude while maintaining roughly the same inference latency. We verify RankMixer's universality with online A/B tests across two core application scenarios (Recommendation and Advertisement). Finally, we launch 1B Dense-Parameters RankMixer for full traffic serving without increasing the serving cost, which improves user active days by 0.3% and total in-app usage duration by 1.08%.

Recently, motivated by the outstanding achievements of diffusion models, the diffusion process has been employed to strengthen representation learning in recommendation systems. Most diffusion-based recommendation models typically utilize standard Gaussian noise in symmetric forward and reverse processes in continuous data space. Nevertheless, the samples derived from recommendation systems inhabit a discrete data space, which is fundamentally different from the continuous one. Moreover, Gaussian noise has the potential to corrupt personalized information within latent representations. In this work, we propose a novel and effective method, named Asymmetric Diffusion Recommendation Model (AsymDiffRec), which learns forward and reverse processes in an asymmetric manner. We define a generalized forward process that simulates the missing features in real-world recommendation samples. The reverse process is then performed in an asymmetric latent feature space. To preserve personalized information within the latent representation, a task-oriented optimization strategy is introduced. In the serving stage, the raw sample with missing features is regarded as a noisy input to generate a denoising and robust representation for the final prediction. By equipping base models with AsymDiffRec, we conduct online A/B tests, achieving improvements of +0.131% and +0.166% in terms of users' active days and app usage duration respectively. Additionally, the extended offline experiments also demonstrate improvements. AsymDiffRec has been implemented in the Douyin Music App.

Retrievers, which form one of the most important recommendation stages, are responsible for efficiently selecting possible positive samples to the later stages under strict latency limitations. Because of this, large-scale systems always rely on approximate calculations and indexes to roughly shrink candidate scale, with a simple ranking model. Most of the existing methods mainly focus on incorporating complicated ranking models. However, index structure is not improved, which also bottlenecks the whole effectiveness. In this paper, we propose a novel index structure: streaming Vector Quantization model, as a new generation of retrieval paradigm. Streaming VQ attaches items with indexes in real time, granting it immediacy. Moreover, through meticulous verification of possible variants, it achieves additional benefits like index balancing and reparability, enabling it to support complicated ranking models as existing approaches. Streaming VQ has been deployed and replaced all major retrievers in Douyin and Douyin Lite, resulting in remarkable user engagement gain.

Sequential recommendation, a critical task in recommendation systems, predicts the next user action based on the understanding of the user's historical behaviors. Conventional studies mainly focus on cross-behavior modeling with self-attention based methods while neglecting comprehensive user interest modeling for more dimensions. In this study, we propose a novel sequential recommendation model, Pyramid Mixer, which leverages the MLP-Mixer architecture to achieve efficient and complete modeling of user interests. Our method learns comprehensive user interests via cross-behavior and cross-feature user sequence modeling. The mixer layers are stacked in a pyramid way for cross-period user temporal interest learning. Through extensive offline and online experiments, we demonstrate the effectiveness and efficiency of our method, and we obtain a +0.106% improvement in user stay duration and a +0.0113% increase in user active days in the online A/B test. The Pyramid Mixer has been successfully deployed on the industrial platform, demonstrating its scalability and impact in real-world applications.

Interest modeling in recommender system has been a constant topic for improving user experience, and typical interest modeling tasks (e.g. multi-interest, long-tail interest and long-term interest) have been investigated in many existing works. However, most of them only consider one interest in isolation, while neglecting their interrelationships. In this paper, we argue that these tasks suffer from a common "interest amnesia" problem, and a solution exists to mitigate it simultaneously. We propose a novel and unified framework in the retrieval stage, "Trinity", to solve interest amnesia problem and improve multiple interest modeling tasks. We construct a real-time clustering system that enables us to project items into enumerable clusters, and calculate statistical interest histograms over these clusters. Based on these histograms, Trinity recognizes underdelivered themes and remains stable when facing emerging hot topics. Its derived retrievers have been deployed on the recommender system of Douyin, significantly improving user experience and retention. We believe that such practical experience can be well generalized to other scenarios.

User preferences follow a dynamic pattern over a day, e.g., at 8 am, a user might prefer to read news, while at 8 pm, they might prefer to watch movies. Time modeling aims to enable recommendation systems to perceive time changes to capture users' dynamic preferences over time, which is an important and challenging problem in recommendation systems. Especially, streaming recommendation systems in the industry, with only available samples of the current moment, present greater challenges for time modeling. There is still a lack of effective time modeling methods for streaming recommendation systems. In this paper, we propose an effective and universal method Interest Clock to perceive time information in recommendation systems. Interest Clock first encodes users' time-aware preferences into a clock (hour-level personalized features) and then uses Gaussian distribution to smooth and aggregate them into the final interest clock embedding according to the current time for the final prediction. By arming base models with Interest Clock, we conduct online A/B tests, obtaining +0.509% and +0.758% improvements on user active days and app duration respectively. Besides, the extended offline experiments show improvements as well. Interest Clock has been deployed on Douyin Music App.

Context modeling for Vision Recognition and Automatic Image Annotation (AIA) has attracted increasing attentions in recent years. For various contextual information and resources, semantic context has been exploited in AIA and brings promising results. However, previous works either casted the problem into structural classification or adopted multi-layer modeling, which suffer from the problems of scalability or model efficiency. In this paper, we propose a novel discriminative Conditional Random Field (CRF) model for semantic context modeling in AIA, which is built over semantic concepts and treats an image as a whole observation without segmentation. Our model captures the interactions between semantic concepts from both semantic level and visual level in an integrated manner. Specifically, we employ graph structure to model contextual relationships between semantic concepts. The potential functions are designed based on linear discriminative models, which enables us to propose a novel decoupled hinge loss function for maximal margin parameter estimation. We train the model by solving a set of independent quadratic programming problems with our derived contextual kernel. The experiments are conducted on commonly used benchmarks: Corel and TRECVID data sets for evaluation. The experimental results show that compared with the state-of-the-art methods, our method achieves significant improvement on annotation performance.