基于聚类与专家信任的协作过滤推荐算法研究

发布时间：2018-07-09 15:25

本文选题：协作过滤 + 用户特征　；参考：《海南大学》2017年硕士论文

【摘要】：随着Web2.0技术的迅速普及,Internet上的数据和资源都处于指数增长阶段,这就会使用户面临信息过载的问题。推荐系统是解决这个问题的有效办法之一,其中协作过滤利用用户评分矩阵,计算用户之间相似度,并根据邻近用户喜好向目标用户进行推荐。但是协作过滤存在新用户冷启动、数据稀疏性、可拓展性等问题。本文针对这些问题,利用混合推荐算法的优点,对协作过滤算法进行了相应的改进,主要工作如下:(1)梳理归纳了推荐系统的相关算法,阐述了国内外学者对协作过滤算法的研究现状。针对协作过滤算法存在的缺点进行深入分析,并探讨如何利用现有技术来提升协作过滤推荐算法的性能。(2)针对冷启动和推荐精度问题,提出了综合用户特征及专家信任的协作过滤推荐算法。通过引入用户特征,利用用户填写的注册信息有效缓解推荐系统中冷启动问题。通过引入专家信任,能够比较用户与专家的相似性,从而计算用户-专家相似度矩阵,进而有效降低了数据集的稀疏性,提高预测的准确度。从实验结果可以看出,该算法能够有效缓解冷启动问题,明显提高了系统的推荐精度。(3)针对数据稀疏性和可拓展性问题,提出了基于奇异值分解与K-means++聚类的协作过滤推荐算法。通过将用户聚成多个簇,然后在与目标用户相似的簇中寻找邻居集,这样可以缓解数据的稀疏性,同时也降低了计算量。通过奇异值分解将用户-项目评分矩阵进行降维,并对稀疏矩阵进行填充,这些模型可在离线的状态下进行建立。从实验结果可以看出,该算法能够有效缓解稀疏性,并提高推荐精度。
[Abstract]:With the rapid popularization of Web2.0 technology, data and resources on Internet are in an exponential growth stage, which will enable users to face the problem of information overload. Recommendation systems are one of the most effective ways to solve this problem, in which collaborative filtering uses the user rating matrix to calculate the similarity between users and to the target based on the preferences of adjacent users. Users are recommended. But collaborative filtering has problems such as new users' cold start, data sparsity and scalability. In this paper, we use the advantages of the hybrid recommendation algorithm to improve the collaborative filtering algorithm. The main work is as follows: (1) the relevant algorithms of the recommendation system are summarized and summarized, and the scholars at home and abroad are expounded. The current research status of collaborative filtering algorithm is discussed. The shortcomings of collaborative filtering algorithm are deeply analyzed, and how to use existing technology to improve the performance of collaborative filtering recommendation algorithm is discussed. (2) aiming at the problem of cold start and recommendation accuracy, a collaborative filtering recommendation algorithm with comprehensive user characteristics and expert trust is proposed. Using the registration information filled by the user effectively alleviates the cold start problem in the recommendation system. By introducing the expert trust, the similarity between the user and the expert can be compared, and the user expert similarity matrix is calculated, and the sparsity of the data set can be reduced effectively and the accuracy of the prediction can be improved. From the experimental results, it can be seen that the algorithm can be used. Effectively alleviating the cold start problem, it obviously improves the recommendation accuracy of the system. (3) in view of the problem of data sparsity and scalability, a collaborative filtering recommendation algorithm based on singular value decomposition and K-means++ clustering is proposed. By clustering the users into multiple clusters and finding the neighbors in the clusters similar to the target users, the data can be alleviated. By the singular value decomposition, the user project score matrix is reduced and the sparse matrix is filled. These models can be set up in the off-line state. It can be seen from the experimental results that the algorithm can effectively alleviate the sparsity and improve the precision of the recommendation.
【学位授予单位】：海南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【相似文献】