基于层级嵌入学习的社交策展内容推荐

Hierarchical embedding-based social curation recommendation

为对社交策展网站中的用户-采集册-采集条目进行统一建模,提出一种基于层级嵌入学习的社交策展内容推荐方法。基于语言词向量嵌入模型,通过隐性层级嵌入学习和显性层级嵌入学习两种策略,学习得到用户、采集册和采集条目在同一低维嵌入空间的表达。通过该方法获得的嵌入表达较好捕捉了用户与内容之间的关联关系,可以准确匹配需求用户和感兴趣内容,应用到个性化信息推荐问题中。在图像策展网站Pinterest上的实验验证了该方法在采集册推荐和采集条目推荐问题上的有效性。

To efficiently model the hierarchical user-pinboard-pin structure in typical social curation websites,a social curation content recommendation method based on hierarchical embedding was proposed.Inspired by the recent progress in word embedding language models,two specific solutions,i.e.,implicit hierarchical embedding and explicit hierarchical embedding were introduced to learn the low-dimension representations for users,pinboards and pins in the same embedded space.The derived embedded representation well captured the correlation between user and pinboards/pins.Accurate connection between desired user and interesting content was thus enabled to facilitate personalized information recommendation.Experimental results on image curation network Pinterest demonstrate that,better novelty as well as precision is achieved in the application of pinboard recommendation and pin recommendation.