个性化推荐中协同过滤算法研究

发布时间：2018-07-17 06:26

【摘要】：随着信息技术和网络技术的迅速发展,人们获取信息的方式越来越多,但网络中信息的爆炸式增长使用户迷失在信息的海洋中,越来越难准确地检索出自己真正需要的信息,即出现了信息超载现象。为了解决该问题,个性化推荐系统应运而生,它不需要用户主动输入任何信息,通过分析用户的历史行为构建用户兴趣模型,从而主动向用户推荐其可能感兴趣的信息。个性化推荐系统的核心是其所采用的推荐算法,在众多推荐算法中,协同过滤推荐算法是研究最多、应用最广的推荐算法。本文详细分析了协同过滤推荐算法的工作流程,并针对数据稀疏问题和兴趣迁移问题提出了一种改进的协同过滤推荐算法来提高推荐系统的推荐质量。本文的主要研究工作：(1)针对评分矩阵数据稀疏性问题,本文提出一种改进的协同过滤算法,首先根据项目属性对整个项目集进行聚类,然后在每个聚类中用Slope One算法进行填充,计算用户相似性时采用的是用户在每个聚类上的加权相似性。(2)传统协同过滤推荐算法依靠评分数据进行推荐,没有考虑到用户兴趣随时间改变这一因素,越早的评分利用价值越低,为了更准确地预测评分,本文将艾宾浩斯遗忘规律引入到推荐过程中,通过为每个评分增加一个时间权重来提高推荐系统的推荐质量。(3)为了降低最近邻居中极少数用户对目标项目评分的影响,先利用目标用户与最近邻居中每个用户的相似性对目标项目进行预测评分,得到一个虚拟最近邻矩阵,然后在虚拟最近邻矩阵中再次利用相似性进行预测。最后,为了验证本文提出的改进算法的有效性,采用MovieLens数据集分别对传统协同过滤推荐算法和本文提出的改进算法进行试验比较,实验结果证明本文提出的改进算法效果更佳。
[Abstract]:With the rapid development of information technology and network technology, there are more and more ways for people to obtain information. However, the explosive growth of information in the network makes users lost in the ocean of information, and it is more and more difficult to accurately retrieve the information they really need. That is, the phenomenon of information overload. In order to solve this problem, personalized recommendation system emerges as the times require, it does not need users to input any information actively, through analyzing the historical behavior of users to build user interest model, thus actively recommend the information that users may be interested in. The core of personalized recommendation system is its recommendation algorithm. Among many recommendation algorithms, collaborative filtering recommendation algorithm is the most widely studied and widely used recommendation algorithm. In this paper, the workflow of collaborative filtering recommendation algorithm is analyzed in detail, and an improved collaborative filtering recommendation algorithm is proposed to improve the recommendation quality of recommendation system. The main work of this paper is as follows: (1) aiming at the sparsity of scoring matrix data, an improved collaborative filtering algorithm is proposed. Firstly, the whole itemset is clustered according to the item attributes. Then, Slope one algorithm is used to fill each cluster, and the weighted similarity of users on each cluster is used to calculate the user similarity. (2) the traditional collaborative filtering recommendation algorithm relies on the score data to recommend. Not taking into account the change of user interest over time, the earlier the score is used, the lower the value. In order to predict the score more accurately, this paper introduces the rule of Ibinhaos forgetting into the recommendation process. By adding a time weight to each score to improve the recommendation quality of the recommendation system. (3) in order to reduce the impact of a very small number of users in the nearest neighbor on the target item score, A virtual nearest neighbor matrix is obtained by using the similarity between the target user and each user in the nearest neighbor, and then the similarity is used to predict the target item again. Finally, in order to verify the effectiveness of the proposed improved algorithm, MovieLens dataset is used to compare the traditional collaborative filtering recommendation algorithm with the improved algorithm proposed in this paper. The experimental results show that the improved algorithm proposed in this paper is more effective.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】