基于用户行为的视频推荐.pdf

摘要 摘要 5G技术的快速发展也带来了信息和互联网技术的飞速发展，他们在丰富人 们生活的同时，也改变了人们的生活方式。以观看视频作为休闲娱乐的用户增多， 加大了市场对视频的需求，同时在这个人人自媒体的时代，向互联网投放视频的 用户也与日俱增。在浩大的互联网资源信息平台中，对于视频信息的个性化推荐 成为各大平台挽留客户的重要手段，在资源库中发掘满足用户需求的视频推荐给 用户，是视频业务领域不可或缺的重要技术。 本文详细介绍了视频推荐领域常用推荐技术，其中基于协同过滤推荐是视频 推荐领域的热门方法，但同时也存在一些问题。传统协同过滤技术忽略了用户与 项目之间的交互信息，缺少针对用户的个性化推荐。因此如何利用视频本身特点， 结合用户的历史行为信息找出他们之间的潜在联系，解决协同过滤中交互信息缺 失这一问题对于向用户进行个性化推荐来说具有极大意义。 针对上述问题及推荐系统中普遍存在的用户评分矩阵数据稀疏性的特点，本 文提出一种基于视频本体特征的推荐方法。首先，构建视频本体特征，并进行视 频本体特征的相似度量；其次，针对用户未评分的视频，对比其与存在历史评分 的视频的本体特征的相似度；再次，综合多部与未观看视频之间本体相似度较高 的看过的视频评分信息，完成用户—视频评分矩阵的缺失数据的评分估计，并对 评分矩阵进行填充；最后，将完善后的矩阵，在基于模型的协同过滤推荐基础上， 选取适合高维矩阵的SVD 模型，搭建用户评分推荐模型，并将Top-N 电影推荐 结果返回用户。通过从视频平台爬取的实际数据，验证改进后的模型在提高用户 评分预测精度上确实有较好的效果，说明基于视频本体特征的推荐模型在一定程 度上确实有效解决了用户—视频评分矩阵数据稀疏性问题，并将用户个性化需求 纳入模型之中。 关键词：视频推荐；推荐系统；矩阵分解；本体相似性 I 目录 目录 第一章 绪论········································································································1 第一节 选题背景及研究意义············································································1 一、选题背景······································································································1 二、研究意义······································································································2 第二节 国内外相关研究动态及文献综述·························································4 一、国外推荐系统相关研究动态············································································4 二、国内推荐系统相关研究动态············································································4 三、视频推荐系统的研究·····················································································5 第三节 主要研究内容·······················································································7 第二章 相关技术研究·······················