基于LDA主题模型的标签混合推荐研究

Purpose/significance]For the current tag recommendation methods' results not satisfied, this paper aims to improve the traditional similarity calculation method and combine a variety of tag recommendation methods to improve the recommended accuracy.[Method/process]Based on the idea of content and collaborative filtering, LDA is used to calculate the similarity then find the neighbor of resources and users, and combine keywords which are extracted from resource contents to construct the tag hybrid recommendation model. Finally, "Douban reading" is taken as an example to verify the model's effectiveness and compared with several tag recommendation methods.[Result/conclusion]In the social tagging system,three dimensions including user, resource and tag should be considered.Only from one single angle will inevitably cause incomplete results.At the same time, the introduction of LDA in similarity calculation can exploit the potential semantic relation and improve the recommended quality. And the combination of a variety of ways to learn from each other can make the results more satisfactory.