信息安全研究 ›› 2019, Vol. 5 ›› Issue (11): 1027-1032.

• 内容安全与人工智能专题 • 上一篇    下一篇

今日头条推荐系统背后的技术原理

刘波   

  1. 今日头条
  • 收稿日期:2019-11-08 出版日期:2019-11-15 发布日期:2019-11-20
  • 通讯作者: 刘波
  • 作者简介:刘波, 今日头条高级研发工程师,主要研究领域为推荐算法。

The Technical Principles Behind the Toutiao Recommendation System

  • Received:2019-11-08 Online:2019-11-15 Published:2019-11-20

摘要: 今日头条自2012年推出以来,致力于通过人工智能连接人与信息,让优质、丰富的信息得到高效、精准的分发,为用户创造价值,目前已成为国内最大的信息平台之一,拥有科技、体育、健康、美食、教育、三农、国风、NBA等超过100个垂直领域,覆盖了图文、图集、小视频、短视频、短内容、直播、小程序等多种信息体裁.为满足内容爆炸时代用户获取有价值信息的需求,今日头条率先在推荐系统领域作了大量探索和实践,将大规模集群流式计算、自然语言处理、深度学习等技术应用到内容推荐和内容质量评估处理、文本和图片识别、视频理解和推荐、多媒体创作和视觉平台等,取得显著成效.

关键词: 推荐系统, 模型, 内容质量, 文本识别, 图片识别, 信息安全

Abstract: Since launched in 2012, Toutiao has been committed to connect people and information by artificial intelligence, make the distribution of high quality, abundant information more efficient and accurate, create values for users. Toutiao has now become one of the largest information platforms in China, which own more than 100 vertical channels, such as science and technology, sports, health, food, education, agriculture, rural areas and farmers, Chinese style, the NBA, etc, supporting multiple sort of genres of online information, such as graphic and text, atlas, small video, short video, short articles, live video streaming, mini program, etc. To meet the users demands for obtaining valuable information in such an exploding information era, Toutiao has taken the lead in the field of recommendation systems and made significant achievements through a lot of explorations and practice, such as applying the largescale cluster streaming computing, natural language processing, deep learning to content recommendation and quality evaluation processing, text and image recognition, video understanding and recommendation, multimedia creation and visual platform, etc.

Key words: recommendation system, model, content quality, text recognition, image recognition, information security