基于自编码器与多模态数据融合的视频推荐方法  

Fusion of auto encoders and multi-modal data based video recommendation method

作　　者：顾秋阳 琚春华[2] 吴功兴[2] GU Qiuyang;JU Chunhua;WU Gongxing(Zhejiang University of Technology,School of Management,Hangzhou 310023,China;Zhejiang Gongshang University,Hangzhou 310018,China)

机构地区：[1]浙江工业大学管理学院,浙江杭州310023 [2]浙江工商大学,浙江杭州310018

出　　处：《电信科学》2021年第2期82-98,共17页Telecommunications Science

基　　金：国家自然科学基金资助项目(No.71571162);浙江省社会科学规划重点课题项目(No.20NDJC10Z);国家社会科学基金应急管理体系建设研究专项(No.20VYJ073);浙江省哲学社会科学重大课题项目(No.20YSXK02ZD)。

摘　　要：现今常用的线性结构视频推荐方法存在推荐结果非个性化、精度低等问题,故开发高精度的个性化视频推荐方法迫在眉睫。提出了一种基于自编码器与多模态数据融合的视频推荐方法,对文本和视觉两种数据模态进行视频推荐。具体来说,所提方法首先使用词袋和TF-IDF方法描述文本数据,然后将所得特征与从视觉数据中提取的深层卷积描述符进行融合,使每个视频文档都获得一个多模态描述符,并利用自编码器构造低维稀疏表示。本文使用3个真实数据集对所提模型进行了实验,结果表明,与单模态推荐方法相比,所提方法推荐性能明显提升,且所提视频推荐方法的性能优于基准方法。Nowadays,the commonly used linear structure video recommendation methods have the problems of non-personalized recommendation results and low accuracy,so it is extremely urgent to develop high-precision personalized video recommendation method.A video recommendation method based on the fusion of autoencoders and multi-modal data was presented.This method fused two data including text and vision for video recommendation.To be specific,the method proposed firstly used bag of words and TF-IDF methods to describe text data,and then fused the obtained features with deep convolutional descriptors extracted from visual data,so that each video document could get a multi-modal descriptors,and constructed low-dimensional sparse representation by autoencoders.Experiments were performed on the proposed model by using three real data sets.The result shows that compared with the single-modal recommendation method,the recommendation results of the proposed method are significantly improved,and the performance is better than the reference method.

关 键 词：自编码器 多模态表示 数据融合 视频推荐 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]