Research On Trust-Based Recommendations In Internet Of Educational Services

The rapid development of Web-based education leads a profound change in educational forms and patterns. The explosive growth of the scale of educational services and users makes the information overload problem more prominent and serious. Some special educational application requirements, which have issued from the new educational models and network structures, bring new challenges to the existing recommendation methods. In this thesis, the core purpose is to meet the trustworthy educational service evaluation and recommendation demand of educational service requesters. From the service oriented research angle, this thesis defines and analyzes the concepts and characteristics of educational service and Internet of Educational Services (IoES). Then, in view of the special application requirements and construction features of three kinds of educational mode of IoES, this thesis uses the trust-and reputation-based research method and technology to put forward some corresponding trustworthy educational service evaluation and recommendation models. The proposed models satisfy the requesters’ educational requirements of service quality and credibility, and perform efficient, personalized recommendations as well. The main research work and contributions are as follows:(1) In IoES where educational services’ quality is implicit, a trustworthy educational service evaluation model is proposed using the evaluation results of the service quality and the provider reputation. Based on the comprehensive analysis of the multidimensional, dynamic, and implicit natures of the educational services’ quality, the model incorporates third-party authority’s quality monitoring and educational service consumers’ quality rating to realize the educational services quality accurately. Educational service providers’ reputation is defined and evaluated from its service quality and service consistency. The proposed trustworthy ES evaluation model integrates the real-time quality and providers’ reputation, and considers the requesters’ personalized requirements of quality as well. Experiments are done under the disturbance of uncertainty factors in the network. The experimental results show that the proposed method can better forecast the changing trustworthy of educational services and verify the stability and applicability of the model.(2) In IoES where the network structure is "decentralization", a recommendation model is proposed based on trust relation between users and recommendation relation between users and educational services. First, a trust network and the recommendation relations between users and services are built. Then, in the trust network which is composed of service trust and recommendation trust, the model integrate the direct service trust, indirect service trust, recommendation trust, and reputation to measure trust value of the service providers to the requester. The corresponding problems such as the trust propagation and aggregation are solved to overcome the sparsity issue. Experiments which are based on a trust/reputation management simulator show that the proposed method is more effective than some other methods in againsting some common malicious behavior of educational service users and has relatively better recommendation effect.(3) In IoES where a question answering (q&a) system is concerned, the trust mining methods based on social relations and behavior are proposed and the algorithm using incorporated trust to perform diverse repliers recommendations is put forward. In q&a systems, users’ rating data is scarce. Using the information of social relationships and behavior, some trust information can be mined from different semantic of cognition. In the q&a community, different kinds of trust has different contributions in promoting questions’ answering. A diverse replier recommendation method is proposed using incorporated trusts and proved to have better recommendation performance. According to the different contexts and application demands, using different trust incorporating strategy to perform different replier recommendations can effectively improve the performance of the recommendation and meet the demand of the corresponding application. Validated by the experiments which are based on real q&a data that different trust and different diverse repliers recommendation methods have different results in promoting questions answering and diverse repliers recommendation method has better recommendation results than compared methods.