摘要
随着互联网产品的多样化,群推荐的应用需求也在不断上升,而目前群推荐技术大多停留在个性化推荐的简单叠加阶段。由于群组成员之间的关系复杂性,提高群推荐的准确性和用户满意度相对于单一用户的推荐更加困难。为尽可能的提高推荐质量,过去的大部分研究考虑到用户之间的相互关系及相似性,但由于群组推荐的情境变化很多,很少有方法可以满足和适用所有的情境。通过使用一种针对群推荐系统特殊性进行改进的新颖数据融合方法,试图根据用户推荐列表排序的相似性研究出用户偏好模式,进而推算出用户在群里的偏好权重。实验方法与几种常见的数据融合方法进行比较,通过比较召回率和F1值对推荐结果进行评价,发现在推荐效果上有一定提高。
The Internet products have been increasingly diverse in recent years, which in some way enlarges the need for group recommendation. However, most applications of group recommendation technology are simply summing up the recommendation results of the members from the given groups. Considering the complexity of the relations among the group members, the precision of the recommendation and the satisfaction of group members would be more unpredictable and uncontrollable to improve than those of a single user. To address this problem, the former researches tend to attach importance to the similarity and the relations among the members. These researches have shown good results. However, they still fail to come up with a universal approach to this task under a diversity of circumstances. This paper provides a new data fusion algorithm focusing on group recommendation systems. The algorithm specifies the preference models among the members based on the similarity of their ranks of recommendations, then calculates the weights of the members to return the group recommendation lists. The algorithm is compared with some basic data fusion algorithms. Then the recall and F1 result upon these algorithms are tested.Finally we find that the algorithm we use in the paper show better results than the other algorithms.
出处
《巢湖学院学报》
2017年第6期27-35,92,共10页
Journal of Chaohu University
基金
高校自然科学研究重点项目(项目编号:KJ2015A400)
安徽省质量工程教学研究重点项目(项目编号:2015jyxm612)
高校优秀青年人才支持计划重点项目(项目编号:gxyq ZD2016423)
安徽省教育厅省级教学研究项目(项目编号:2016jyxm0160)
关键词
群推荐
协同过滤
数据融合
排序相似性
偏好模式
group recommendation
collaborative filtering
data fusion
rank similarity
preference model
