基于向量化标签的视频推荐算法研究与实现

Research and implementation of video recommendation algorithm based on vectorized label

当前,推荐算法的基础数据质量或算法本身等原因,导致复杂兴趣推荐困难、解释性差、推荐效果不佳等问题出现。标签是反映用户兴趣的重要信息,文章基于标签的视频推荐,从相似度计算、用户画像、多路召回、排序模型等方面展开讨论。通过电影和电视剧推荐实验发现,访客数(UV)、访问量(PV)、人均播放量、人均播放时长等指标均取得一定的提升,同时增加了推荐的可解释性。

At present,the basic data quality of the recommendation algorithm or the algorithm itself and other reasons,resulting in complex interest recommendation difficulties,poor interpretation,poor recommendation effect and other problems.Label is an important information to reflect the interests of users.The article is based on label video recommendation,which is discussed from the aspects of similarity calculation,user portrait,multi-channel recall,sorting model and so on.Through the film and TV series recommendation experiment,it is found that the number of visitors(UV),the number of visits(PV),the per capita broadcast volume,the per capita broadcast time and other indicators have been improved,and the explanatory of the recommendation has been increased.