基于聚类的加权Slope One推荐技术研究

发布时间：2019-02-16 10:27

【摘要】：互联网中信息规模的爆炸式增长,满足了用户对信息的需求。但庞大的信息量使得用户难以快速定位到有用信息,降低了信息的使用率,导致了信息过载问题的出现。个性化推荐技术是一种面向用户进行个性化推荐的有效手段,其核心是推荐算法。Slope One算法是一种简单而高效的基于项目的协同过滤算法,能够在少量数据情况下达到较好的推荐效果,已经得到了广泛应用。但现有的Slope One算法无法在数据稀疏情况下做精确推荐,评分过程中会利用无关项目预测评分且无法快速感知用户兴趣的变化。为了解决上述问题,本文对权重的计算方法加以改进,提出改进的加权Slope One算法,再引入数据挖掘的相关技术,对数据进行分类和预处理,提出基于聚类的加权Slope One算法。所做的主要工作如下:第一,在传统的K-Means算法的基础上,提出一种自动生成K个聚类中心的基于最小生成树的K-Means算法,有效解决传统的K-Means算法因初始聚类中心选取的随机性引起的局部最优问题,提高聚类效果;第二,利用聚类结果对原始项目评分矩阵进行预测填充,解决算法存在的稀疏性问题,并根据聚类结果缩小推荐候选集的规模,减少推荐算法计算量;第三,考虑项目属性和项目评分对项目相似度影响程度的不同,引入项目属性和项目评分的项目综合相似度计算方法,提高项目相似度的准确性;第四,为了在算法中更好的反应用户兴趣的变化,突出新数据作用削弱旧数据。在推荐算法中加入时间权重,考虑影响时间权重的因素,提出加入访问频率的时间权重函数;第五,根据本文提出的改进算法,设计推荐系统,介绍系统中模块组成、模块间调用关系和模块内部算法流程,利用MovieLens数据集在系统上进行验证。实验证明,基于聚类的加权Slope One算法与传统推荐算法相比,聚类算法的加入能够有效解决稀疏性问题,减少计算量;项目相似度和时间权重的加入提高了算法预测的准确性和时间敏感度。整体算法在平均绝对误差上有着明显的降低,能够有效提高推荐系统整体性能。
[Abstract]:The explosive growth of the scale of information in the Internet meets the needs of users for information. However, the huge amount of information makes it difficult for users to locate useful information quickly, reduce the utilization rate of information, and lead to the problem of information overload. Personalized recommendation technology is an effective way for users to make personalized recommendation. Its core is that the recommendation algorithm. Slope One algorithm is a simple and efficient collaborative filtering algorithm based on project. It has been widely used to achieve good recommendation effect in a small amount of data. However, the existing Slope One algorithm can not make accurate recommendation in the case of sparse data. In the process of evaluation, independent items are used to predict the score and the changes of user interest can not be quickly perceived. In order to solve the above problems, this paper improves the method of weight calculation, proposes an improved weighted Slope One algorithm, introduces the related technology of data mining, classifies and preprocesses the data, and proposes a weighted Slope One algorithm based on clustering. The main works are as follows: first, based on the traditional K-Means algorithm, a K-Means algorithm based on minimum spanning tree is proposed to generate K clustering centers automatically. In order to improve the clustering effect, the traditional K-Means algorithm can solve the local optimal problem caused by the randomness of the initial clustering center selection. Secondly, the original item scoring matrix is predicted and filled with the clustering results to solve the sparse problem of the algorithm. According to the clustering results, the size of the recommended candidate set is reduced, and the calculation amount of the recommendation algorithm is reduced. Thirdly, considering the difference between item attribute and item score on project similarity, we introduce the method of project attribute and item score to calculate the project similarity, and improve the accuracy of project similarity. Fourth, in order to better reflect the change of user interest in the algorithm, highlight the role of new data weakening the old data. The time weight is added to the recommendation algorithm, and the time weight function of the access frequency is put forward considering the factors that affect the time weight. Fifthly, according to the improved algorithm proposed in this paper, we design the recommendation system, introduce the module composition, the call relationship between modules and the algorithm flow inside the module, and use the MovieLens data set to verify the system. The experiments show that compared with the traditional recommendation algorithm, the weighted Slope One algorithm based on clustering can effectively solve the sparse problem and reduce the computational complexity. The addition of item similarity and time weight improves the accuracy and time sensitivity of the algorithm. The overall algorithm can significantly reduce the average absolute error and can effectively improve the overall performance of the recommendation system.
【学位授予单位】：北京工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】