Abstract: With the maturity of Web services related standards and technologies, Web services recommendation based on Quality of Service（QoS） plays a decisive role in user experience. How to accurately predict the Qos value is a hot research topic today. In the past, collaborative filtering algorithms based on neighbors or models use two-dimensional information of “user-service”. The predicted QoS value is static and not accurate. The dimension of time information is introduced into the tensor model, and the three-dimensional tensor of “user service time” can make the QoS prediction more consistent with the user’s requirements. The Bayesian method is used to solve the tensor decomposition. The introduction of probabilistic interpretation and analysis of the system provides a Bayesian inference framework for priori probabilities, which improves the accuracy of QoS prediction. Experiments show that the prediction results using this algorithm have smaller mean absolute deviation than other algorithms, and solve the problem of data sparsity.

Key words: Web services, Quality of Service（QoS） prediction, tensor decomposition, Bayesian algorithm

摘要： 随着Web服务相关标准和技术的日趋成熟，基于服务质量（QoS）的Web服务推荐对用户体验起着决定性作用。如何准确预测Qos值是当今的研究热点。以往基于近邻或模型的协同过滤算法，采用的是“用户-服务”二维信息，预测的QoS值是静态的且精准性不高。将时间信息维度引入张量模型，建立“用户-服务-时间”的三维张量可使QoS预测值更加符合用户需求特点，用贝叶斯方法求解张量分解，引入概率意义下对于系统的解释和分析，提供一套先验概率引入先验知识的贝叶斯推断框架，提高了QoS预测的精确度。实验表明，使用该算法的预测结果较其他算法相比较有更小的平均绝对误差，很好地解决了数据稀疏度问题。

关键词: Web服务, 服务质量（QoS）预测, 张量分解, 贝叶斯算法