【作者】 李昕；

【导师】 王永贵；

【作者基本信息】 辽宁工程技术大学 ， 软件工程（专业学位）， 2022， 硕士

【摘要】 目前,各推荐系统普遍面临数据稀疏的问题,仅将用户与项目的联系简单地停留在信息表层,并未深度挖掘用户对不同项目的兴趣程度,推荐准确率较低,影响推荐效果。针对上述问题,本文提出了一种融合模糊聚类和改进相似度的协同过滤推荐算法,在推荐技术中应用聚类思想的同时构建一种新的相似度计算方法,以解决推荐技术中存在的问题。首先,从项目的角度出发,在数据处理的过程中,基于项目的协同过滤算法充分挖掘项目间的数据关系,填充原始矩阵的零值以进行后续的推荐过程,直接降低了数据稀疏性;其次,从用户的角度出发,使用模糊C-均值聚类算法对用户进行聚类,并借助狼群算法全局搜索的优势优化初始聚类中心,增加用户的聚类效果。再次,以用户属性为基础构建用户画像,计算得到不同用户的标签化指数,使用改进后的相似度计算公式计算、筛选得到最近邻居集合。最后,根据上述过程中得到的结果,对用户未产生交互的项目计算预测评分后进行最终的推荐。在真实的数据集上进行独立实验以及对比实验,结果表明,本文所提改进后的算法实现了一定的突破,数据稀疏性降低了20%左右,缓解了数据稀疏的问题;平均绝对误差、均方根误差均小于对比试验,可证明本文所提算法的有效性。该论文有图29幅,表21个,参考文献62篇。更多还原

【Abstract】 The research status showed that data sparsity was still the biggest problem faced by recommendation systems,currently.Only the connection between users and projects simply stayed on the information surface,and didn’t deeply mine users’ interest in different projects in the information field.The recommendation accuracy was low,which affected the recommendation effect.To solve the above problems,this paper proposes a collaborative filtering recommendation algorithm integrating fuzzy clustering and improved similarity.While applying the clustering idea in the recommendation technology,a new similarity calculation method is constructed to alleviate the problems existing in the recommendation technology.Firstly,from the perspective of project,the Item-based Collaborative filtering algorithm fully excavated the data relationship between projects and filled the zero of the original matrix for the subsequent recommendation process,which directly reduced the data sparsity;Secondly,from the perspective of users,fuzzy c-means clustering algorithm was used to cluster users,and with the advantage of global search of wolf swarm algorithm,the initial clustering center was optimized to increase the clustering effect of users.Then,different users were labeled based on user attributes,the labeling indexes of different users were calculated,and the nearest neighbor set was calculated and filtered by using the improved similarity calculation formula.Finally,according to the results obtained in the above process,the final recommendation was made after calculating the prediction score for the items that didn’t generate interaction.Independent experiments and comparative experiments were carried out on real data sets.The results showed that the improved algorithm had achieved a certain breakthrough,the data sparsity was reduced by about 20%,and the problem of data sparsity was alleviated;The mean absolute error and root mean square error were less than the comparative test,which can prove the effectiveness of the proposed algorithm.更多还原