Research On Big Service Discovery

With the rapid development of personal computer, Internet and telecommunication, especially for the cloud computing, Web 2.0 and mobile computing, the cost of service development, deploy-ment, management, access and maintenance largely decrease, followed by explosive growth of services. Meanwhile, the communication data across user and user, user and service, service and service progressively grows. For instance, the invocation records of Google API service are more than 10 billion per day. The era of big data for service computing has already been in our daily life.Under the big service data circumstance, how to effectively utilize the service relevant data to discover the complex service is becoming the hot topic in both academia and industry. Traditional service computing research can not satisfy the requirement of service discovery in new environment for its limitations in scalability data. In this work, we focus on four key research topics of massive complex service discovery, namely service search, service recommendation, service selection and service management. The main contributions of this work could be summarized as follows.Firstly, to tackle the low performance of service matching in traditional service search sce-nario, we introduce the service tagging data to improve the search performance. Specifically, we propose a distributed sparse technique to connect tags and existing service description data. Partic-ularly, due to the inherent properties of service tagging data, e.g., fuzziness, extensiveness, or even viciousness, we propose tag classification and recommendation approaches to improve the quality of service tagging data. Experiments based on real data demonstrate the effectiveness of service tagging data and the proposed approaches.Secondly, to improve the overall quality of service recommendation, we introduce a personal service recommendation framework that utilizes user location information. Typically, we propose a time smooth strategy to re-balance the importance of historic data. Meanwhile we incorporate the user location information to extend the power of matrix factorization model which is the pow-erful recommendation framework in industry. We also propose to evaluate our framework on a massive real world dataset. And the experimental report shows the effectiveness of our proposed approaches.Thirdly, to handle the problem of missing QoS data in service selection scenario, we propose a novel service selection mechanism by predicting QoS data in a collaborative filtering manner. Then we propose a EPCC strategy to remove the difference among services and a extended K-means method to group users into a neighborhood, which directly improves the quality of service selection task. We also use accelerated method to boost the calculation process. We also propose to evaluate our framework on a massive real world dataset. And the experimental report shows the effectiveness of our proposed approaches.Last but not least, to build a bridge between academia and industry on the service QoS man-agement system, we propose a methodology a unified framework to guide designers and architects by performing a massive quantitative analysis on QoS data. We then propose various measurement functions and solvers that affect different aspects of service QoS management system. Also we distill lessons on the key factors and interfaces for further usage. We also propose to evaluate our framework on a massive real world dataset. And the experimental report shows the effectiveness of our proposed approaches.