Abstract:With the increase of content pages and interactive buttons in online services such as online-shopping and video-watching websites, industrial-scale recommender systems face challenges in multi-domain and multi-task recommendations. The core of multi-task and multi-domain recommendation is to accurately capture user interests in multiple scenarios given multiple user behaviors. In this paper, we propose a plug-and-play \textit{\textbf{P}arameter and \textbf{E}mbedding \textbf{P}ersonalized \textbf{Net}work (\textbf{PEPNet})} for multi-domain and multi-task recommendation. PEPNet takes personalized prior information as input and dynamically scales the bottom-level Embedding and top-level DNN hidden units through gate mechanisms. \textit{Embedding Personalized Network (EPNet)} performs personalized selection on Embedding to fuse features with different importance for different users in multiple domains. \textit{Parameter Personalized Network (PPNet)} executes personalized modification on DNN parameters to balance targets with different sparsity for different users in multiple tasks. We have made a series of special engineering optimizations combining the Kuaishou training framework and the online deployment environment. By infusing personalized selection of Embedding and personalized modification of DNN parameters, PEPNet tailored to the interests of each individual obtains significant performance gains, with online improvements exceeding 1\% in multiple task metrics across multiple domains. We have deployed PEPNet in Kuaishou apps, serving over 300 million users every day.

What problem does this paper attempt to address?

The paper primarily addresses the challenges faced by multi-domain and multi-task recommendation systems in online services (such as online shopping and video streaming websites) and proposes a method called PEPNet (Parameter and Embedding Personalized Network). Specifically, PEPNet aims to solve the following key issues: 1. **Multi-domain and Multi-task Recommendation**: With the increase in content pages and interactive buttons on online service platforms, recommendation systems need to handle data from multiple domains and capture users' diverse behavioral preferences in different scenarios. The paper points out that accurately capturing users' interests in different scenarios becomes a core challenge in this context. 2. **Enhanced Personalization in Recommendations**: Traditional recommendation models often focus on a single prediction task in a single domain (e.g., click-through rate (CTR) prediction). However, in practical applications, recommendation needs are dispersed across different scenarios. To better utilize user feedback and provide a better user experience, recommendation systems need to capture users' diverse behavioral preferences for different targets. 3. **Addressing the Incomplete Double Seesaw Phenomenon**: Due to the overlap of users and items across different domains and the dependencies between different tasks, simply training separate models for each task or domain is not only costly but also fails to fully leverage the commonalities in the data, potentially leading to suboptimal performance. Moreover, directly modeling all data together may ignore the differences between domains and tasks. This phenomenon is referred to by the authors as the "incomplete double seesaw phenomenon." To address the above issues, PEPNet solves the problems through the following approaches: - **Dynamic Adjustment of Embedding Layers and Deep Neural Network Hidden Units**: PEPNet receives personalized prior information as input and dynamically adjusts the underlying embedding layers and the hidden units of the top-level deep neural network through a gating mechanism. This allows the model to select features of varying importance across different users and domains. - **Personalized Embedding Network (EPNet) and Parameter Personalized Network (PPNet)**: EPNet is responsible for personalized selection of embeddings across multiple domains to integrate features of varying importance for different users; PPNet, on the other hand, personalizes the parameters of the deep neural network across multiple tasks to balance different users' preferences for targets with varying sparsity. - **Plug-in Architecture**: PEPNet is designed as an efficient, low-cost, and easily deployable plug-in network that can be integrated into existing models. Through these methods, PEPNet aims to improve the performance of recommendation systems, especially in complex multi-domain and multi-task scenarios. Experimental results show that PEPNet achieves significant performance improvements on an industrial-grade short video dataset.

PEPNet: Parameter and Embedding Personalized Network for Infusing with Personalized Prior Information

A Multi-Task Embedding Based Personalized POI Recommendation Method

Self-Attention Mechanism Enhanced User Interests Modeling for Personalized Recommendation Services in Cyber-Physical-Social Systems.

Personalized Elastic Embedding Learning for On-Device Recommendation

Neural graph personalized ranking for Top-N Recommendation

Personal Recommendation Using Deep Recurrent Neural Networks In Netease

A Personalized Explainable Learner Implicit Friend Recommendation Method

An Efficient Adaptive Transfer Neural Network for Social-aware Recommendation

Dynamic Parameterized Network for CTR Prediction

Parallel Knowledge Enhancement Based Framework for Multi-behavior Recommendation

PROPN: Personalized Probabilistic Strategic Parameter Optimization in Recommendations

Denoising Pre-Training and Customized Prompt Learning for Efficient Multi-Behavior Sequential Recommendation

PC-IEN: a click-through rate prediction method based on dynamic collaborative personalized interest extraction

Heterogeneous Fusion of Semantic and Collaborative Information for Visually-Aware Food Recommendation

Progressive Layered Extraction (PLE): A Novel Multi-Task Learning (MTL) Model for Personalized Recommendations

EdgeDIPN

Metapath-guided Heterogeneous Graph Neural Network for Intent Recommendation

Personalizing Intervened Network for Long-tailed Sequential User Behavior Modeling

Personalized Recommendation Via Inductive Spatiotemporal Graph Neural Network

DPAN: Dynamic Preference-based and Attribute-aware Network for Relevant Recommendations

Perceive Your Users in Depth: Learning Universal User Representations from Multiple E-commerce Tasks