Abstract: There are many enterprise and business resources in the multi-service value chain.Through recommendation technology,high-quality business resources are recommended to different enterprise users to help enterprises reduce costs and increase value.But the traditional recommendation technology has the problems of business resource reduction and the long tail effect,which seriously affects the recommendation quality.To solve the above problems,a recommendation algorithm of business resources for multi-service value chain was proposed.The algorithm learned the correspondence between the optimal clustering algorithm and the meta-attributes of the data set through meta learning to select the optimal clustering algorithm for the data set at a certain time and convert the original high-dimensional scoring matrix of business resources into sub-scoring matrices of multiple low-dimensional.In the sub-scoring matrix,the penalty factor optimization collaborative filtering was used to alleviate the long tail problem of business resource recommendation.Though experimental results confirmed,the effectiveness of the proposed algorithm was proved.Compared with other recommended algorithms,the proposed algorithm had significantly improved in mean absolute error,root mean square error,comprehensive evaluation index and coverage.

Key words: multi-service value chain, business resource recommendation, clustering algorithm, penalty factor, meta-learning

摘要： 多服务价值链中有很多企业和业务资源。通过推荐技术,将优质的业务资源推荐给不同企业用户,帮助企业降本增值。但传统的推荐技术存在业务资源简约和长尾问题,严重影响推荐质量。为解决上述问题,提出一种面向多服务价值链的业务资源推荐算法。该算法首先通过元学习学习最优聚类算法和数据集元属性之间的对应关系,以便可以为某时刻的数据集选择最优聚类算法,将高维业务资源评分矩阵转化为多个低维度的子评分矩阵。然后在子评分矩阵中通过惩罚因子优化协同过滤来缓解业务资源推荐长尾问题。最后通过实验研究证实,所提算法相比于其他推荐算法,在平均绝对误差、均方根误差、综合评价指标以及覆盖率上分别有明显的提升,证明了该算法的有效性。

关键词: 多服务价值链, 业务资源推荐, 聚类算法, 惩罚因子, 元学习