论文部分内容阅读
在传统协同推荐方法中,相似性的度量是整个方法的核心.在数据稀疏情况下,现有相似度计算方法仅使用历史评分数据,难以准确反映用户之间的相似程度;相关改进方法在考虑用户共同评分数量对相似度的影响时,引入的重叠度参数需要手动调整,限制了方法实用性.针对上述问题,本文提出一种基于最近邻用户重排序(DRNN)的相似度方法,充分利用项目类别信息,根据不同的目标项目动态调整邻居集内用户权重,能更准确地刻画用户之间的相似性;并提出修正的重叠度因子弥补现有方法中手动调整参数的不足,增强了方法实用性.实验结果表明,该方法可以明显提升预测结果的准确性.
In the traditional collaborative recommendation method, the similarity measure is the core of the whole method.In the case of data sparseness, the existing similarity calculation method only uses the historical scoring data, and it is difficult to accurately reflect the similarity degree between users; when the related improvement method is considered When the influence of the common scoring quantity on the similarity, the introduced overlap parameter needs to be manually adjusted, which limits the practicability of the method.Aiming at the above problem, this paper proposes a similarity approach based on the nearest neighbor user reordering (DRNN) Project category information, dynamically adjust the user weights in the neighbor set according to different target items, and more accurately describe the similarity between users; and propose a modified overlap factor to make up for the lack of manual adjustment parameters in the existing methods and enhance the method Practicality The experimental results show that this method can significantly improve the accuracy of the prediction results.