Classic collaborative filtering algorithm is memory-based non-parametric local model, the model is applied K-nearest neighbors ( KNN) technology to target user neighbor recommended items for the target preferences, as a recommendation to the user for effective standards. The method in predicting a longer computing time required, and restrictions on certain parameters , we can not guarantee that all users effectively predict. In order to solve the above problem this paper presents a new model based on clustering recommended method. The algorithm first assume that the target user and recommend items to a certain degree of probability can be classified as a different user mode and recommend key mode;Individual users by calculating the recommended mode of entry mode for each rating , as well as the user the probability of belonging to different user-mode, recommended mode of entry of the probability of belonging to different projects; Resulting in the target user for specific items recommended prediction score. With the classic collaborative filtering recommendation algorithm comparison of the results,changing the program can be predicted in a short time to all users for all recommended items score , and its effects with other methods recommend a good contrast with the improvements.%经典的协作式过滤算法基于记忆的非参数局部模型,该模型应用最近邻算法(K-nearest neighbors,KNN)技术,把目标用户近邻对于目标推荐项的喜好,作为向该用户进行有效推荐的标准.该方法在预测时需要较长的运算时间,并且在特定参数的限制下,不能保证对所有的用户进行有效预测.为了解决以上问题,介绍1种基于聚类模式的新的推荐方法.该算法首先假设目标用户和推荐项均能以一定的概率划归于不同的用户模式和推荐项模式中;通过计算各个用户模式对于各个推荐项模式的评分,以及用户属于不同用户模式的概率,推荐项属于不同项目模式的概率;从而产生目标用户对于具体推荐项的预测评分.通过与经典的协作式过滤推荐算法结果的对比,该方案可以在较短的时间预测所有用户对于所有推荐项的评分,并且其推荐效果与其他方法对比有了很好的改进.
展开▼
