Privacy-enhanced personalization for online learning

点击翻译按钮获取中文摘要

Computer systems that personalize their actions in reaction to personal preferences or requirements of human users do so by building profiles of the users. These profiles are based on knowledge gained either from analysis of user interactions with the system or from outside data sources. In either circumstance, there is a potential threat to user privacy. Research by me and others has been devoted to providing personalization support while preserving user privacy. One key approach has been to separate the system's knowledge of the personal identity of the user from the behaviour that is being monitored. This can be achieved by providing a pseudonym for the user and associating behaviour information with the pseudonymous identity. But pseudonymity introduces both social and technical challenges for personalization systems.  In computer-based learning environments, preserving learner privacy is important in situations where the release of personal information may result in inappropriate stereotypes or discrimination, potential harassment by other learners, or academic dishonesty through identity theft. There is a trade off between protecting learnerprivacy and gaining necessary and sufficient knowledge for a learning support system to work effectively. In larger learning communities where personal trust relationships may be less pervasive (andsometimes ill-advised), a mechanism for protecting user privacy and gauging trust in others is needed. Methods of pseudo-identity can be successfully applied here. For many learners, particularly children, greaterautomated protection is appropriate. Methods to identify and curb risky behaviour on the part of learners are needed.  Techniques for protecting privacy and modelling trust (appropriate for the learning domain) will be investigated in my research. This includes developing secure and reliable models for privacy-enhanced personalization through identity management, models for private reputation transfer from one pseudonymous identity to another, and models for managing multiple identities associated with a single learner.