基于联邦学习框架的制造服务个性化推荐方法研究

Personalized Manufacturing Service Recommendation Method Based on Federated Learning Framework

针对工业云平台中海量的制造服务使得用户难以进行服务选择,以及推荐算法的使用使得平台存在泄露用户数据风险的问题,提出一种基于联邦学习框架的制造服务个性化推荐方法。采用基于模型的协同过滤推荐算法,在矩阵分解中考虑用户和服务的评分偏置项以提高推荐准确度。推荐算法所用的数据被分类存储于用户端与服务器端,其中包含用户隐私的数据留存至用户本地。本地数据经加密算法进行脱敏后,与服务器端的数据结合以共同训练模型。最终,用户对制造服务的不同属性的评分被预测得出。根据制造服务质量的特点,在多属性预测评分的基础上进一步量化用户个性化倾向。通过个性化倾向因子对预测评分的干预得到服务的Pareto非支配解集,生成制造服务的推荐列表。实例验证表明,该推荐方法能够在保护用户隐私的前提下获得良好的预测准确度,在考虑用户个性化偏好的情况下保证推荐服务的预测评分均值较高,从而帮助用户进行制造服务选择。

To solve the problem of the massive manufacturing services in the industrial cloud platform make it difficult for users to select services and the use of recommendation algorithm makes the platform have the risk of divulging user data,a personalized recommendation method of manufacturing services based on Federated Learning framework is proposed.The model-based Collaborative Filtering recommendation algorithm is adopted,and the scoring bias of users and services is considered in the matrix decomposition to improve the recommendation accuracy.Data used in the recommendation algorithm is classified and stored in the client and server,of which data containing users'privacy is retained locally.Local data is combined with data in the server to train the model after being desensitized by the encryption algorithm.Finally,users'scores on different attributes in manufacturing service are predicted.Users’personalization tendency is further quantified by multi-attribute prediction scores according to the characteristics of manufacturing service quality.Pareto non-dominant solution set of the service is obtained through the intervention of the personalization tendency factor on the prediction score,and the recommended list of the manufacturing service is generated.The experiment shows that the recommendation method can obtain good prediction accuracy without compromising users’privacy and the average prediction score of the recommendation service is high with users’personalized preferences being considered,thus helping users better choose manufacturing services.