Research On Personalized Web Service Recommendation Based On Collaborative Filtering And QoS

The rising development of internet technology, web service selection andrecommendation is becoming an important research problem attracting great attentionsfrom both academia and industry researchers. With the number increasing of Webservices, recommending and selecting optimal Web services for users has become oneof the most challenging issues in the field of service computing. In the presence ofmultiple Web services with conform or similar functionalities, both functional andnonfunctional Quality-of-Service (QoS) attributes should be taken into account to helpuser selecting the most suitable services according to their needs. QoS of Web servicescan change at run time due to various reasons (e.g., server workload, network condition,etc.). Consequently, methods where Web services are statically evaluated areinappropriate. Instead, a new approach is needed, in which runtime changes in the QoSof the services are taken into account. To address these problems, this thesis proposesseveral methods based on Collaborative Filtering (CF) and QoS for personalized webservice recommendation. To overcome the drawbacks described above, differencebetween services and context are introduced in Web service recommedation for betterrecommendation accuracy. Furthermore, two novel hybrid collaborative filteringalgorithms are proposed to solve the problem of weights between different collaborativefiltering methods. The main research and contributions of this thesis is described asfollowing:①An overview of current development and problems of Web services is given.And the main research areas, special characteristics and critical challenges of Webservices are also summarized. Then research progress on Web services technology issummarized and classified, which introduces the significance of research in this thesisand provides the correlative theory for further research.②A brief introduction to collaborative filtering technology is presented. And thedevelopment of collaborative filtering is also analyzed, which includes the importantacademic and commercial significance. Furthermore, the main technology ofcollaborative filtering is summarized, in which the characteristics and applicationscenarios is also analyzed. These analyses have set up the theoretical foundation forresearch on personalized web services recommendation by QoS and collaborative filtering.③The collaborative filtering is introduced in web services recommendation areato construct a personalized web services recommendation system based on CF and QoSfor addressing the problems of dynamic, personalization and context-free. According touser-service matrix of QoS values, the missing QoS values of web services can beobtained by CF method. The predicted QoS values will be ranked according to certainrules, and then the web services candidate with optimal predicted QoS performance canbe recommended to users. Firstly, a web services recommendation method viadifference of services is proposed, and the difference of services is first introduced inpersonalized web services recommendation using real-world QoS dataset, theexperimental results demonstrate that the difference between services is moreappropriate to present the relationship between services than similarity and thedifference based method provides better prediction accuracy than other similarity basedmethod. Secondly, to address the problem of how to identify the weights betweendifferent CF methods in hybrid web service recommendation, two novel hybridrecommendation methods of web services based on neural network are first proposed, inwhich BP and RBF neural networks are employed to train weights among different CFmethods, respectively. These hybrid methods combine the advantages of different CFmethods, and experiments show that our hybrid methods achieve better predictionaccuracy and time complexity than the state-of-the art method WSRec. Finally, toovercome the context-free problem in web services recommendation, a personalizedcontext-aware recommendation method for web services is proposed. The experimentspresent that the context of users and web services have great influence on performanceof web services recommendation, and the context-aware method can provide bothefficient and effective recommendation.④Comprehensive experiments are conducted employing real-world Web serviceQoS dataset to present the recommendation performance of proposed methods in thisthesis. The large-scale real-world dataset includes1.5millions Web service invocations,and these web services are randomly selected100Web services from totally21,197publicly available Web services to be invoked by service users in more than20countries. This dataset is the largest one among the published work of the real-worldQoS data set.⑤This thesis is under the background of national natural science foundation "The behavior monitoring and dependable evolution of large distributed software system ".According to the predicted QoS values provided by our personalized web servicerecommendation methods based on CF and QoS, research on the changes of QoS forweb services in distributed systems, then the evolution of distributed systems can bedriven. The system of QoS driven web services evolution is constructed, and thearchitecture, functional modules are also provided.