考虑时间动态性的协同过滤算法及其应用研究

发布时间：2017-12-31 04:03

本文关键词：考虑时间动态性的协同过滤算法及其应用研究　出处：《河北工业大学》2015年硕士论文　论文类型：学位论文

【摘要】：随着电子商务的崛起,越来越多的人选择使用电商去购物,同时对电商的购物体验有个更高的要求。个性化推荐系统就是为了增加客户的购物体验而生的,在现实中应用广泛。个性化推荐算法中,应用最成功的是协同过滤推荐算法。本文分析了现有协同过滤算法存在的不足,以及关于时间动态性的研究现状,指出了一个重要的问题——商品是有先后购买顺序的。传统的协同过滤算法只考虑了用户或商品之间的相关程度,并没有根据商品的序列模式进行挖掘并推荐。本文结合现有算法的优点,加入了时间动态性的特点,提出了一个考虑时间动态性的协同过滤推荐综合性改进算法,并通过真实的电子商务销售数据进行实验测试,验证了其推荐效果更佳。本文的主要内容包括以下几点:(1)通过对现有文献的综述研究,找到目前学者们的研究重点和研究方向,总结现有研究的特点,分析传统协同过滤推荐算法的不足之处,总结出现有算法大多只在某一两个方面进行改进,而加入时间动态性的研究也只停留在用户兴趣的改变上,从而引出了本文要研究的方向和重点。(2)结合现有研究中的优秀改进方向和改进方法,加入商品序列模式的筛选和去除热门效应的影响,提出一个综合性的协同过滤推荐算法解决方案。其核心的思路可以总结为首先对原始数据集进行预处理,包括去噪音和聚类等;然后用一个新的相似度函数进行计算最近邻居,这个相似度计算函数考虑了热门系数,评分系数和时间系数;最后使用GSP算法挖掘商品的序列模式,对推荐结果集进行再次筛选,解决了用户买鼠标后给其推荐电脑的问题。(3)收集了亚马逊的真实评论数据,对本文提出的改进算法进行实验设计和推荐效果验证。与传统的协同过滤算法进行比较,提高了推荐的准确率,并改善了传统算法中存在的数据稀疏性等问题,验证了改进算法的推荐效果更优秀。
[Abstract]:With the rise of e-commerce, more and more people choose to use electricity to go shopping, the electricity supplier shopping experience has a higher request. Personalized recommendation system is to increase the customer's shopping experience, widely used in reality. The personalized recommendation algorithm, the most successful application of the collaborative filtering algorithm. This paper analyzes the shortcomings of the existing collaborative filtering algorithm, and the research status on time dynamic, pointed out an important issue, commodities have successively purchase order. The traditional collaborative filtering algorithms only consider the degree of correlation between users or commodity, and not according to the sequence pattern of commodity by mining and recommended. This paper combines the advantages of existing algorithm, joined the time dynamic characteristics, a new collaborative filtering algorithm recommended time dynamic comprehensive, and E-commerce sales data real test, verify its better recommendation effect. The main contents of this paper include the following: (1) through a review of existing literature, find the current research emphases and directions of scholars, the characteristics of existing research, analysis on the deficiency of the traditional collaborative filtering recommendation algorithm the present algorithms are only in a certain improvement in one or two aspects, and adding time of dynamic research only stay in the user interest change, which leads to the direction and the focus of this study. (2) combined with the excellent improvement direction in current research and improvement methods, influence the selection of goods series the hot and removal effect, put forward a comprehensive collaborative filtering algorithm solutions. Its core idea can be summed up as the first of the original data set is preprocessed, including To remove the noise and clustering; then calculate the nearest neighbor with a new similarity function, the similarity function is considered hot coefficient, score coefficient and time coefficient; finally we use the GSP algorithm for sequential pattern mining commodities, on the recommendation result set again screening, to solve the user to buy the mouse after its recommended computer problems. (3) collected real data verify the Amazon review, experimental design and effect of improved recommendation algorithm is proposed in this paper. Compared with the traditional collaborative filtering algorithm, improves the accuracy of recommendation, and improve the data sparseness problem of traditional algorithm, verify that the improved algorithm is more effective good.

【学位授予单位】：河北工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.3;F713.36

【相似文献】