基于位置聚类和张量分解的Web服务推荐算法

Web service recommendation based on location clustering and tensor decomposition

基于服务质量(Qo S)的Web服务推荐能在众多功能相似的Web服务中发现最能满足用户非功能需求的Web服务,但Qo S属性值预测算法仍存在预测准确度不高和数据稀疏性的问题。针对以上问题,提出了一种基于位置聚类和分层张量分解的Qo S预测算法Clust TD,该算法基于用户和服务的位置属性将用户和服务聚类成多个局部组,分别对局部组和全局的用户、服务和时间上下文进行张量建模和分解,将局部和全局张量分解的Qo S预测值进行加权组合,同时考虑了局部和全局因素,获得最终Qo S预测值。实验结果表明,该算法具有较高的Qo S预测准确率和Web服务推荐质量,并能在一定程度上解决数据稀疏性问题。

Web service recommendation based on Quality-of-Service(QoS)is of vital importance for users to find theproper Web service among huge numbers of functionally similar Web services. But current QoS predicting algorithms stillhave the data sparse problem and cannot predict the QoS values accurately. Focusing on these problems, a method calledClustTD is proposed in this paper. This method is based on location clustering and hierarchical tensor decomposition.Firstly, it clusters users and services into several local groups based on their location, and models local and global triadictensors respectively for the relations of user -service-time. Then the hierarchical tensor decomposition is performed on thelocal and global triadic tensors. Finally, the predicted QoS value by local and global tensor decomposition is combined asthe missing QoS values. Comprehensive experiment shows that this method achieves higher prediction accuracy and recommendingquality of Web service, and can partly solve the problem of data sparse.