Dynamically Predicting the Quality of Service: Batch, Online, and Hybrid Algorithms

Ya Chen,Zhong-An Jiang

doi:10.1155/2017/9547869

Abstract

This paper studies the problem of dynamically modeling the quality of web service. The philosophy of designing practical web service recommender systems is delivered in this paper. A general system architecture for such systems continuously collects the user-service invocation records and includes both an online training module and an offline training module for quality prediction. In addition, we introduce matrix factorization-based online and offline training algorithms based on the gradient descent algorithms and demonstrate the fitness of this online/offline algorithm framework to the proposed architecture. The superiority of the proposed model is confirmed by empirical studies on a real-life quality of web service data set and comparisons with existing web service recommendation algorithms.

Highlights

The quality of service or QoS has been exploited by many application domains as an importance metric for users to evaluate the quality of provided services
We evaluate the performance of our proposed algorithm (TSVD), matrix factorization (MF), and TF on the collected data sets
We propose a hybrid system designing framework for dynamic service quality prediction, which takes

Summary

Introduction

The quality of service or QoS has been exploited by many application domains as an importance metric for users to evaluate the quality of provided services. Existing recommendation approaches have been applied in the web service discovering domain, for example, userbased collaborative filtering (CF) [14], item-base CF [15], and hybrid CF [16] These web service and API prediction systems assume that all user-service invocation records are collected and the trained model will not change with the accumulation. The basic idea is to use a tensor factorization-based approach to evaluate timeaware personalized QoS values These studies achieve performance improvements in comparison with traditional models; the temporal dynamics of user-service invocation records is not considered. We design a hybrid model which jointly combines the online and offline algorithms Such system is capable of monitoring the change of QoS values and adapting itself in real-time to the new arriving data.

Architecture

Time-Dependent Quality of Web Service Prediction Model

Algorithm

Experimental Evaluation

Conclusion