LinkSage：基于GNN的Pinterest站外内容理解

Adopted by Pinterest multiple user facing surfaces, Ads, and Board.

被Pinterest多个用户界面、广告和画板采用。

Jianjin Dong | Staff Machine Learning Engineer, Content Quality; Michal Giemza| Machine Learning Engineer, Content Quality; Qinglong Zeng | Senior Engineering Manager, Content Quality; Andrey Gusev | Director, Content Quality; Yangyi Lu | Machine Learning Engineer, Home Feed; Han Sun | Staff Machine Learning Engineer, Ads Conversion Modeling; William Zhao | Software Engineer, Boards Foundation, Jay Ma | Machine Learning Engineer, Ads Lightweight Ranking

LinkSage: Graph Neural Network based model for Pinterest off-site content semantic embeddings

LinkSage：基于图神经网络的Pinterest站外内容语义嵌入模型

Background

背景

Pinterest is the visual inspiration platform where Pinners come to search, save, and shop the best ideas in the world for all of life’s moments. Most of the Pins are linked to off-site content to provide Pinners with inspiration and actionability. It is critical to understand off-site content (images, text, structure), because understanding their semantics is an important factor in assessing how safe (e.g. community guidelines), functional, relevant, and actionable (e.g. Ads and Shopping) the off-site content is. More importantly, Pinterest can have a better understanding of Pinterest users through users’ click through events. Both of the above can improve overall engagement and monetization of Pinterest contents. To achieve it, we developed LinkSage, which is a Graph Neural Network (GNN) based model that learns the semantics of landing page contents.

Pinterest是一个视觉灵感平台，用户可以在这里搜索、保...