面向IPTV的混合式自适应推荐系统关键技术研究与实现
发布时间：2018-10-23 10:41
【摘要】： IPTV作为新一代有线数字电视产品,自从进入中国以来,用户量增长迅速。根据权威机构IDC的预测,到2009年底,中国IPTV用户量将达到460万,而到2013年,这数字将增长到1310万,并将进入一个井喷式的发展。 IPTV的主要优势在于其良好的互动性。通过IPTV,用户将在“IP机顶盒+电视机”上告别单一被动的节目接收,走向更为丰富多彩的互动数字娱乐生活。内容服务提供商可以在IPTV上提供大量高质量的数字图像、视频、音频、游戏、远程教育、广告等内容。在这种环境下,大量的信息容易让用户产生信息迷失。因此为用户提供精准高质的个性化服务成为一种迫切的需求。目前世界范围内对个性化服务的研究主要归为对推荐系统的研究范畴。 文章首先深入分析现有推荐系统算法所存在的不足,其中包括新用户问题以及混合式过滤算法所采用的固定混合比造成的推荐质量下降等问题。作者针对这些问题展开研究讨论。 首先针对新用户问题,文章提出了基于人口属性的协作过滤算法,这个算法将人口属性信息相似度引入协作过滤算法,并和PCC计算所得相似度进行混合得到新的相似度。采用这个相似度计算最近邻并产生推荐。实验分析表明,文章提出的基于人口属性的协作过滤在用户评分稀少,用户profile稀疏的时候能够有效提高推荐质量。 之后针对传统混合式推荐系统造成推荐质量下降问题,提出了基于递度下降的混合式自适应推荐算法。本算法引入自学习机制,让系统自动调整混合式推荐系统的混合比。实验表明,这个算法在一定程度上提高了推荐精度,并且不增加过多的额外计算时间。 文章的第三个成果是通过对IPTV平台特性的分析,以及将它同现有以个人电脑为终端的推荐系统的比较,总结出面向IPTV的推荐系统所应该具有的特性:用户零学习成本、用户零额外操作。针对这个特性,文章设计了一个用户喜好挖掘算法,通过分析用户的访问日志,自动获取用户的喜好。经过系统一年的线上运行,证明此算法运行效果良好。
[Abstract]:IPTV as a new generation of cable digital television products, since entering China, the number of users growing rapidly. According to IDC, an authoritative organization, the number of IPTV users in China will reach 4.6 million by the end of 2009, and will increase to 13.1 million by 2013. And will enter a blowout development. The main advantage of IPTV lies in its good interactivity. IPTV, users will bid farewell to the single passive program reception on "IP set-top box TV" and move towards a more colorful interactive digital entertainment life. Content service providers can provide high-quality digital images, video, audio, games, distance education, advertising and other content on IPTV. In this environment, a large number of information is easy to make users lose information. Therefore, it is an urgent need to provide users with accurate and high quality personalized services. At present, the research on personalized service is classified as recommendation system. Firstly, this paper deeply analyzes the shortcomings of the existing recommendation system algorithms, including the problem of new users and the degradation of recommendation quality caused by the fixed mixing ratio of hybrid filtering algorithm. The author studies and discusses these problems. To solve the problem of new users, a collaborative filtering algorithm based on population attributes is proposed in this paper. This algorithm introduces the similarity of population attributes into the collaborative filtering algorithm, and mixes with the similarity calculated by PCC to obtain a new similarity. This similarity is used to calculate the nearest neighbor and produce recommendations. The experimental results show that the proposed collaborative filtering based on population attributes can effectively improve the recommendation quality when the user score is scarce and the user profile is sparse. After that, a hybrid adaptive recommendation algorithm based on transitivity reduction is proposed to solve the problem of the deterioration of recommendation quality caused by the traditional hybrid recommendation system. This algorithm introduces self-learning mechanism and allows the system to automatically adjust the hybrid ratio of hybrid recommendation systems. Experiments show that the proposed algorithm improves the recommendation accuracy to some extent and does not increase the extra computation time. The third result of this paper is to analyze the characteristics of IPTV platform and compare it with the existing recommendation system with personal computer as the terminal, and summarize the characteristics of the recommendation system to IPTV: user zero learning cost. User zero extra operation. Aiming at this feature, this paper designs a user preference mining algorithm, which automatically acquires user preferences by analyzing the user's access log. It is proved that the algorithm works well after one year's running.
【学位授予单位】：华东师范大学
【学位级别】：硕士
【学位授予年份】：2010
【分类号】：TN949.2
本文编号：2288994
[Abstract]:IPTV as a new generation of cable digital television products, since entering China, the number of users growing rapidly. According to IDC, an authoritative organization, the number of IPTV users in China will reach 4.6 million by the end of 2009, and will increase to 13.1 million by 2013. And will enter a blowout development. The main advantage of IPTV lies in its good interactivity. IPTV, users will bid farewell to the single passive program reception on "IP set-top box TV" and move towards a more colorful interactive digital entertainment life. Content service providers can provide high-quality digital images, video, audio, games, distance education, advertising and other content on IPTV. In this environment, a large number of information is easy to make users lose information. Therefore, it is an urgent need to provide users with accurate and high quality personalized services. At present, the research on personalized service is classified as recommendation system. Firstly, this paper deeply analyzes the shortcomings of the existing recommendation system algorithms, including the problem of new users and the degradation of recommendation quality caused by the fixed mixing ratio of hybrid filtering algorithm. The author studies and discusses these problems. To solve the problem of new users, a collaborative filtering algorithm based on population attributes is proposed in this paper. This algorithm introduces the similarity of population attributes into the collaborative filtering algorithm, and mixes with the similarity calculated by PCC to obtain a new similarity. This similarity is used to calculate the nearest neighbor and produce recommendations. The experimental results show that the proposed collaborative filtering based on population attributes can effectively improve the recommendation quality when the user score is scarce and the user profile is sparse. After that, a hybrid adaptive recommendation algorithm based on transitivity reduction is proposed to solve the problem of the deterioration of recommendation quality caused by the traditional hybrid recommendation system. This algorithm introduces self-learning mechanism and allows the system to automatically adjust the hybrid ratio of hybrid recommendation systems. Experiments show that the proposed algorithm improves the recommendation accuracy to some extent and does not increase the extra computation time. The third result of this paper is to analyze the characteristics of IPTV platform and compare it with the existing recommendation system with personal computer as the terminal, and summarize the characteristics of the recommendation system to IPTV: user zero learning cost. User zero extra operation. Aiming at this feature, this paper designs a user preference mining algorithm, which automatically acquires user preferences by analyzing the user's access log. It is proved that the algorithm works well after one year's running.
【学位授予单位】：华东师范大学
【学位级别】：硕士
【学位授予年份】：2010
【分类号】：TN949.2
【参考文献】
相关期刊论文 前2条
1 曾春,邢春晓,周立柱;个性化服务技术综述[J];软件学报;2002年10期
2 曾春,邢春晓,周立柱;基于内容过滤的个性化搜索算法[J];软件学报;2003年05期
，本文编号：2288994
本文链接：https://www.wllwen.com/wenyilunwen/guanggaoshejilunwen/2288994.html
