Abstract: 【Purpose/significance】User Profile can describe the individual and group behavior characteristics of video users in a more
profound manner, and provide reference for personalized recommendation service of video.【Method/process】Through text mining to
the crawling video, users and their viewing data analysis, this paper builds a single user profile. Furthermore, based on K-Means and
LDA model, this paper clusters users and extracts topics, and then mines the characteristics of group users. Video interest tags based
on user profile and time exponential decay combined with video preference and collaborative filtering, are used for video recommenda⁃
tion.【Result/conclusion】Considering the personalized recommendation of time exponential decay, the system's awareness of user in⁃
terest is improved. Combined with video preference and collaborative filtering, the recommended video rating is 0.87, which helps in⁃
crease user loyalty and activity to the site.【Innovation/limitation】Based on the results of text mining of user-generated content, indi⁃
vidual and group user profile are carried out. And then this paper innovatively uses time exponential decay to construct user video in⁃
terest tags to capture dynamic user interest. Due to the limitation of Network Crawler, the amount of experimental data has some limita⁃
tions, and the range of interest in feature extraction is limited.