基于相似性的优化推荐算法研究与设计

发布时间：2018-03-08 15:59

本文选题：个性化推荐　切入点：协同过滤　出处：《华中科技大学》2016年硕士论文　论文类型：学位论文

【摘要】：在互联网时代,个性化推荐系统得到广泛应用。在推荐系统中,推荐算法起着决定性的作用,而协同过滤算法为最为常用的一种推荐算法,采用基于用户的协同过滤对稀疏评分矩阵进行预填充,然后使用基于项目的协同过滤,对未知评分进行预测,可以在一定程度上提高预测精度。但该算法存在两个问题:在预填充阶段,由于用户相似性计算过程受到评分数据稀疏的影响,导致得到的用户最近邻集不够准确,那么利用用户最近邻集填充得到的稠密数据精度较低;在预测阶段,由于度量项目相似性方式不合理,导致得到的项目最近邻集不够准确,最终导致算法精度下降。为了进一步提高该推荐算法的预测精度,针对该算法存在的两个问题做出优化:在预填充阶段,提出了相似性矩阵最优化模型,得到表征用户相似性的矩阵,从而得到更准确的用户最近邻集,该过程受数据稀疏性影响较小,然后根据所得的用户最近邻集对原始评分矩阵预填充,得到精度较高的稠密评分矩阵;在预测阶段,将共评用户与目标用户之间的相似性考虑在内,优化了项目间相似性的度量方式,得到了更准确的项目最近邻集,然后根据所得的项目最近邻集在稠密的评分矩阵基础上预测未知项目的评分。基于上述两点优化策略设计优化算法,最后利用开源数据集进行多组对比试验,实验结果表明,优化后的算法相比优化前的算法在评分预测上有更高的预测精度,从而验证了优化算法的有效性。
[Abstract]:In the Internet era, personalized recommendation system is widely used. In recommendation system, recommendation algorithm plays a decisive role, and collaborative filtering algorithm is one of the most commonly used recommendation algorithms. The sparse score matrix is pre-filled with user-based collaborative filtering, and then the unknown score is predicted by item-based collaborative filtering. The prediction accuracy can be improved to a certain extent, but there are two problems in this algorithm: in the prefill stage, due to the influence of sparse score data on the process of user similarity calculation, the nearest neighbor set is not accurate enough. Then the dense data filled with user nearest neighbor set is of low precision. In the prediction stage, because of the unreasonable method of measuring the similarity of items, the nearest neighbor set of items is not accurate enough. In order to further improve the prediction accuracy of the proposed algorithm, two problems of the algorithm are optimized: in the phase of pre-filling, the optimization model of similarity matrix is proposed. A matrix representing user similarity is obtained, and a more accurate user nearest neighbor set is obtained. The process is less affected by data sparsity, and then the original score matrix is pre-filled according to the user nearest neighbor set. The dense scoring matrix with high accuracy is obtained. In the prediction stage, the similarity between the users and the target users is taken into account, and the measurement method of the similarity between the items is optimized, and a more accurate set of nearest neighbors is obtained. Then, based on the dense score matrix, the evaluation of unknown items is predicted on the basis of the nearest neighbor set. The optimization algorithm is designed based on the above two optimization strategies. Finally, an open source data set is used to carry out multi-group comparative experiments. The experimental results show that, Compared with the algorithm before optimization, the optimized algorithm has higher prediction accuracy, which verifies the effectiveness of the optimization algorithm.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【参考文献】