Dissertation
Dissertation > Industrial Technology > Radio electronics, telecommunications technology > Communicate > Communication network > General issues >

Research on User Preference Based Personalized Service Adaptation Decision Mechanism

Author LiGuoQiang
Tutor YangFangChun
School Beijing University of Posts and Telecommunications
Course Computer Science and Technology
Keywords Personalization Service adaptation User preference Context reasoning Fuzzy evidence theory Learning Classiner System GBML
CLC TN915.09
Type PhD thesis
Year 2012
Downloads 230
Quotes 0
Download Dissertation

Along with the continuous development of computation, communication and network technologies, the future carrying convergence service is transiting from "network centric" and "content centric" to "user centric" mode. And personalization, self adaptation and context awareness will be the main features of the future convergence service besides of ubiquity and convergence. While the personalized service must adapt according to the changing user preference which is caused by the ever-changing context, in other words adjust in accordance with the user preference.But up to now, the intelligent network only implements a limited intelligence by the way of the splitting of service control and service undertaking, even though the publishing of the Parlay/Parlay X framework indeed simplifies the service development and deployment, but the intelligence of the current service is still restricted. It can only adapt in light of the preprogrammed execution flow, and does not have an automatic learning ability.Therefore, considering the characteristics of the convergence ubiquitious network environment the indepth research of the user preference based service adaptation offers a necessary theory basis for "user centric" service implementation. Aiming to resolve the above problems, this article proposed a context dependent preference model as the personalized adaption foundation, which used fuzzy ontology and fuzzy D-S theory to represent and reason the fuzzy context. Then it presented a modified fuzzy XCS algorithm to build this model and make the decision to adapt to the changing environment using this preference model. Besides this article also investigated several mechanism to accelerate the convergence speed of the proposed algorithm, and a coevolutionary mechanism based approach to deal with the slow learning speed and unexpected context adaptation decision of one single user is also presented. Specifically,1. In order to resolve the situation that the current user preference model cannot process the context dependent user preference very well, expecially the fuzzy context based user preference, this paper proposed a context dependent user preference model which combines a rule based representation and quantitative preference, which also models the context parts of the user preference model using fuzzy ontology. Besides, for the purpose of getting around the involved uncertainty reasoning of the fuzzy ontology, this article proposed to use the fuzzy D-S theory to do the uncertainty context infering. This model is used as the basis of the user preference based personalized adaptation and our subsequent studies. 2. With regard to the current status that the service adaptation rules have to be generated manually or learned off line using a machine learning based approach, this paper put forward to use a modified fuzzy XCS (eXtend Classifier System) algorithm to generate these rules. As to the situation that the current XCS algorithm only considers discrete contexts, this paper proposed a modified iXCS algorithm to support the hybrid context types. Besides in order to deal with the current situation that XCS algorithm only takes too simple action formats into account, this article presented iXCSCAP algorithm which has a mechanism combined layered action model and greedy strategy to handle the complex action and parameters problem. Actually iXCSCAP is an extension of iXCS algorithm, which is used to settle the simple action and complex action based adaptation problems seperately. Finally the simulation result revealed that the understandbility and effectiveness of iXCS algorithm comparing to the other learning mechanisms, and the iXCSCAP algorithm also has a much faster convergence speed when facing the complex actions.3. Aim at solving the convergence speed problem of the traditional XCS algorithm as well as iXCS, iXCSCAP algorithms, this article proposed the embedding of the user knowledge, the supervising mechanism of the Fuzzy SUCS (Semi-supervised UCS) algorithm and the simulated metropolis criterion to accelerate its converging speed. The simulated metropolis criterion can speed up the convergence process without users’feedback, and the embedding of the user knowledge mechanism could quicken the leaning speed by manually inserting of user preference rules, while the supervising mechanism of the Fuzzy SUCS (Semi-supervised UCS) algorithm based approach can contribute to the convergence speed without user interactions. These three kinds of methods is applied for different senerios, and experiments conducted in this papers proved theired usability and effectiveness.4. As to the slow training and convergence speed and too many unexpected context decision conditions issue, the collaborative coordination and cooperative coevolution mechanism based XCS algorithm is proposed in this thesis to get assistance of all the similar users. This cooperative coevolutionary mechanism makes use of both the static and dynamic user simility measuring methods, and let the other similar users help to initiate new user’s classifier population as well as cooperate with each other to predict the corresponding action of the uncovered context states. The simulation experiments proved the usability and effectiveness of the presented coevolutionary XCS algorithm.5. Finally in order to push forward the realistic application of cognitive service, a context aware service adaptation infrastructure is presented with the support of the switch and intelligent control research center of state key laboratory to which the author of this thesis belongs, and the core of the framework is a GBML based middleware.

Related Dissertations
More Dissertations