基于遗忘函数和用户的协同过滤推荐算法研究

发布时间：2018-03-11 12:01

本文选题：协同过滤　切入点：遗忘函数　出处：《浙江工业大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着网络的发展,在音乐、购物、阅读、电影等方面,我们更加依赖网络带来的便利。随着用户数目的增加、新兴产品的出现,导致了系统所需存储的数据量日益增加。在“数据过载”时代,如果能够从大量数据中获取有价值的信息,那么就能为用户提供更好的服务。因此在电子商务、社交网络、视频网站等平台上,个性化推荐技术的应用受到了很多关注。目前应用最为广泛的个性化推荐技术是协同过滤推荐技术。协同过滤推荐是通过分析用户的历史行为记录得到用户可能感兴趣的商品,从而为用户进行的推荐。考虑到用户的兴趣存在改变现象,而目前关于兴趣迁移的协同过滤推荐都无法根据用户的实际遗忘过程对项目赋予合适的权重,因此推荐也就不准确。本文采用将艾宾浩斯遗忘曲线拟合后得到的非线性遗忘函数来描述用户的兴趣迁移,对此提出基于遗忘函数和用户的协同过滤推荐改进算法,即在计算相似度时,加入遗忘函数因子来体现用户兴趣的变化。采用改进算法在电影数据集MovieLens和EachMovie中分别进行实验及评估,实验结果表明本文提出的改进算法在两个不同的电影相关数据集中均能提高推荐准确性。因此,采用拟合遗忘曲线后的遗忘函数描述用户兴趣迁移可以为视频网站的影片推荐提供可参考的改进方向。
[Abstract]:With the development of the Internet, in music, shopping, reading, movies etc, we rely more on the convenience brought by the Internet. With the number of users increases, new products, led to the required amount of data storage is increasing. In the "data overload" era, if we can obtain valuable information from large amounts of data so, can provide better service for users. So in the e-commerce, social networking, video website platform, the application of personalized recommendation technology has attracted much attention. The most widely used is the personalized recommendation technology collaborative filtering technology. Collaborative filtering recommendation is to get the user may be interested in goods through history the behavior analysis of users, and for users to recommend. Taking into account the user interest change phenomenon, and the current interest of collaborative filtering recommendation migration to root According to the user's actual process of forgetting to project to give the appropriate weight, therefore recommended is not accurate. This paper applies nonlinear Ebbinghaus forgetting curve fitting the forgetting function to describe the interest migration users, puts forward the forgetting function and user collaborative filtering recommendation algorithm based on the similarity calculation, namely, change add to forget function factor to reflect the user's interest. The improved algorithm experiments and evaluation of MovieLens and EachMovie in the movie data set, the experimental results show that the improved algorithm proposed in this paper focuses on two different film related data can improve the recommendation accuracy. Therefore, by fitting the forgetting curve after forgetting function description of user interest migration recommendation can provide reference for the improvement direction of the video site video.

【学位授予单位】：浙江工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】