摘要
针对视频推荐系统的稀疏性问题、冷启动问题与灰羊用户问题,提出一种基于视频内容检测的协同过滤视频推荐系统。首先,为视频包含的人体动作进行建模并提取时空兴趣点,使用梯度描述符计算兴趣点的主要移动方向与尺度;然后,为每个视频序列构建一个梯度向量的矩阵,使用聚类算法从梯度矩阵中选择初始化向量与代表向量;最终,通过对受欢迎视频作者与用户的评分分析,获得灰羊用户的分组。仿真实验的结果表明,该视频推荐系统较好地解决了视频推荐系统的稀疏性问题、冷启动问题与灰羊用户问题,获得了较高的推荐准确率。
Conceming the sparsity problem, cold start problem and grey sheep user problem of video recommendation systems, a collaborative filtering recommendation system based on video content detection is proposed. Firstly, the human actions in the videos are modeled and space time interest points （STIP） are abstracted, the gradient description is adopted to compute the main motion direction and scale of the space time interest points; then, a gradient vector matrix is constructed for each video sequence, both of the initial vectors and the representation vectors are selected from the gradient matrices by clustering algorithms; lastly, the popular video authors and the users are analyzed about the scores, and the grey sheep users are grouped. Simulation experimental results show that the proposed video recommendation system resolves the sparsity problem, cold start problem and grey sheep user problem efficiently, and realizes a good recommendation accuracy.
出处
《控制工程》
CSCD
北大核心
2018年第2期305-312,共8页
Control Engineering of China
基金
广东省自然科学基金项目(2014A030310380、2015A030310257)
东莞职业技术学院科研基金项目(项目编号2017C01)
关键词
推荐系统
视频内容检测
数据挖掘
协同过滤
机器视觉
Recommendation system
video content detection
data mining
collaborative filtering
machine vision
