融合巴氏系数与综合相似度的改进加权Slope One算法
Weighted Slope One algorithm combining Bhattacharyya coefficient and comprehensive similarity
王文丰 1周雨虹 2周波 3韩佳 2韩龙哲 2董芳 2赵阳2
作者信息
- 1. 南昌工程学院信息工程学院,江西南昌 330099;南昌工程学院江西省水信息协同感知与智能处理重点实验室,江西南昌 330099
- 2. 南昌工程学院信息工程学院,江西南昌 330099
- 3. 江西省鄱阳湖水利枢纽建设办公室,江西南昌 330009
- 折叠
摘要
针对传统加权Slope One算法因过度依赖用户共同评分项导致在过稀疏数据集中预测准确度低的问题,提出一种融合巴氏系数与综合相似度的改进加权Slope One算法(BS-WSO).首先,引入巴氏系数和用户行为偏好对用户相似度计算方法进行改进,依此筛选出待预测的近邻集合;其次,为了优化预测评分,利用巴氏系数和项目流行度计算项目相似度,并将其作为权重因子融入评分计算;最后,将BS-WSO与几种代表性算法进行比较,仿真实验结果表明,BS-WSO算法能有效克服数据过稀疏情况下预测准确度低的缺陷,提高推荐精确度.
Abstract
To solve the problem that the traditional weighted Slope One algorithm relies too much on users'common scoring items,which lead to low prediction accuracy in sparse datasets,an improved weighted Slope One algorithm combining Bab-bitt coefficient and comprehensive similarity is proposed.Firstly,the calculation method of user similarity is improved by in-troducing Bhattacharyya coefficient and user behavior preference,according to which the nearest neighbor set to be predicted is selected;Secondly,in order to optimize the prediction score,the project similarity is calculated by using Bhattacharyya co-efficient and project popularity,and it is incorporated into the score calculation as a weight factor.Finally,the proposed algo-rithm is compared with several representative algorithms under different nearest neighbor numbers.The simulation results show that the proposed algorithm can effectively overcome the defect of low prediction accuracy in the case of too sparse datasets and improve the recommendation accuracy.
关键词
协同过滤/数据稀疏性/相似度/用户偏好Key words
collaborative filtering/data sparsity/similarity/user preference引用本文复制引用
基金项目
国家自然科学基金资助项目(61962036)
国家自然科学基金资助项目(61561035)
江西省水利厅科技重点项目(202325ZDKT17)
出版年
2024
国家科技期刊平台
NSTL
万方数据