个性化混合推荐算法的研究

发布时间：2018-04-13 02:08

本文选题：推荐算法 + 协同过滤　；参考：《浙江大学》2013年硕士论文

【摘要】：网络技术的迅速发展和互联网应用的逐渐丰富,使人们的生活更加的方便多彩,但同时也给人们带来了信息过载的问题。因为互联网上的数据过于庞大,人们没有合适的信息筛选工具,将花费大量时间才能在大规模的信息资源中找到相关的资料。传统的搜索引擎系统在一定程度上解决了这个问题,但同时也存在着一些缺陷：反馈的信息过于杂乱,对用户的操作要求过高以及无法提供个性化的反馈信息等。推荐系统根据用户的长期兴趣特点和行为模式,向用户推荐其感兴趣的信息资源,从而进一步地解决了信息过载的问题。本文的研究目的是为了提升推荐算法的预测准确性。在本文中,我们认为社会主流趋势的影响造成了推荐中的全局特性,而兴趣类似的用户之间的相互影响产生了推荐中的局部特性。然后我们对推荐系统中的全局和局部特性进行了深入的讨论,并且分析了用户评分行为和经典的推荐算法中存在的全局和局部特性。通过分析以后,我们认为全局性的推荐算法和局部性的推荐算法的基本假设和侧重点不同,并且每个用户的行为也是具有明显的差异性,所以将这两类算法进行混合有利于提高预测准确率。于是我们提出了一种个性化混合推荐模型,基于每个用户的行为特征个性化地整合全局性和局部性的推荐算法的结果。最后,我们设计了详细的算法验证性实验,测试数据集是在学术界被广泛采用的推荐测试数据集Movielens。实验表明,个性化混合推荐模型的预测准确率比经典的推荐算法有了大幅度的提高,从而证明了该算法的有效性。
[Abstract]:The rapid development of network technology and the gradual enrichment of Internet application make people's life more convenient and colorful, but also bring people the problem of information overload.Because the data on the Internet is too large, people do not have the appropriate information screening tools, and it will take a lot of time to find the relevant information in the large-scale information resources.The traditional search engine system solves this problem to a certain extent, but at the same time there are some defects: the feedback information is too messy, the operation requirement of users is too high, and the personalized feedback information can not be provided.According to the characteristics of users' long-term interests and behavior patterns, the recommendation system recommends the information resources of interest to the users, thus further solving the problem of information overload.The purpose of this paper is to improve the prediction accuracy of the recommendation algorithm.In this paper, we think that the influence of social mainstream trend results in the global characteristics of recommendation, while the interaction among users of similar interest produces the local characteristic of recommendation.Then we discuss the global and local characteristics of the recommendation system and analyze the behavior of users and the global and local characteristics of the classical recommendation algorithms.After analysis, we think that the basic assumptions and emphases of the global recommendation algorithm and the local recommendation algorithm are different, and the behavior of each user is obviously different.Therefore, mixing the two kinds of algorithms is beneficial to improve the prediction accuracy.So we propose a personalized hybrid recommendation model which integrates the results of global and local recommendation algorithms individualized based on each user's behavioral characteristics.Finally, we design a detailed algorithm validation experiment, the test data set is widely used in the academic community of the recommended test data set Movielens.The experimental results show that the prediction accuracy of the personalized hybrid recommendation model is much higher than that of the classical recommendation algorithm, which proves the effectiveness of the proposed algorithm.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.3

【相似文献】