一种用于视频推荐的基于LDA的深度学习模型  

A LDA-based Topic Attribute-aware in-Depth Learning Model for Video Recommendation

作　　者：程涛 崔宗敏[1] 喻静 CHENG Tao;CUI Zong-min;YU Jing(School of Information Science and Technology,Jiujiang University,Jiujiang 332005,China)

机构地区：[1]九江学院信息科学与技术学院,江西九江332005

出　　处：《计算机技术与发展》2020年第8期86-90,共5页Computer Technology and Development

基　　金：国家自然科学基金(61762055)。

摘　　要：随着视频产业的发展,大量的视频已沉积在信息海洋中,为了缓解这种现象,越来越多的推荐算法开始应用于个性化视频推荐。然而目前的推荐算法都是以协同过滤为自主,都只注重通过评分矩阵提高捕捉用户与视频的低阶的交互,忽略了用户兴趣与视频属性的高阶关联。在这种背景下,文中通过LDA主题模型预测用户兴趣主题,引用干扰词典和关键词典来提高LDA模型对视频文本聚类的准确率,然后利用最近提出的神经协同框架建模用户兴趣和视频属性的高阶关联,把LDA模型与深度学习相结合提出了一种新的模型LIVR用于视频推荐,最后在通过网络爬虫爬取的数据集上验证并对实验结果进行分析。结果表明,该模型的Top-N推荐准确率较常见的几个深度学习模型提高了约1.9个百分点。With the development of video industry,a large amount of videos have been deposited in the information ocean.In order to alleviate this phenomenon,more and more recommendation algorithms have begun to be applied to personalized video recommendation.However,the current recommendation algorithms are all based on collaborative filtering.They only focus on improving the low-level interaction between users and videos through the scoring matrix,ignoring the high-order association between user interests and video attributes.In this context,we use the LDA topic model to predict user interest topics,cite interference dictionaries and key dictionaries to improve the accuracy of LDA model clustering of video texts,and then use the recently proposed neural collaboration framework to model high-order associations of user interests and video attributes.Combined with LDA model and deep learning,a new model LIVR is proposed for video recommendation.Finally,it is verified on the dataset crawled by the web crawler and the experimental results are analyzed.The results show that the Top-N recommendation accuracy of the model is about 1.9 percentage points higher than the common deep learning models.

关 键 词：LDA主题模型 深度学习 隐式反馈 视频推荐 个性化推荐 

分 类 号：TP391.5[自动化与计算机技术—计算机应用技术]