| 摘要 |
When interacting and talking with each other, humans are quite successful at conveying information and reacting appropriately. That is, they are able to interpret situational information which in most cases is only implicitly given. This ability is in stark contrast to most of the current state-of-the-art computer and telecommunications systems, which in large are unaware of this context information. Consequently, there has recently been great interest in making applications more context-aware so that they can adapt to different situations and be more receptive to user's needs. The purpose of this project was to explore how one through examining different kinds of context information can find ways to automatically set a user's presence status. This presence status is useful as it allows for a person to decide whether it is appropriate to initiate communication or not. Further, it was the mission of this project to review how a policy-based approach could be use together with the mentioned aggregated presence status to increase user personalisation of communication. Finally, the project set out to show how one can model and implement such a context-aware system using Ericsson's service creation framework; ServiceFrame. The project used a model-driven approach (MDA) in which most of the effort went into modelling the functionality of the system using formal UML2.0 models and then transform these into code using code generation tools and MDA viewpoints. During the modelling of the system, it was decided to examine user's context information and from this classify the user's aggregated presence status by utilising a supervised learning classifier. In a proof-of-concept solution, the probabilistic classifier, Naïve Bayes, proved to be an extremely fast and accurate classifier which should be considered as a strong option in any future implementations of the system. Personalisation of communications by design was achieved both through allowing completely personal classification rules, as well as through a policy approach. This policy approach consisted of letting a user construct a personal prioritising of his or hers terminals based on his or her aggregated presence status. It is suggested to expand this approach by also taking the group membership of the entity which requests the user information into account. Using supervised learning makes sure the system is not to be solely dependent on any particular context data sensor but rather having the ability to adapt to an ever-changing environment. With the expected explosion of context sensors in areas such as ubiquitous computing [15], this ability to adapt will become increasingly important in future context-aware systems. In addition to this flexibility, the use of supervised learning also introduces the ability to classify completely unknown scenarios. This means that even if the classifier encounters a new scenario which is has never seen before, it tries to predict a status based on the underlying previous data. It stated that these predictions often have a high degree of accuracy with the right training data. The system used the agent oriented architecture, which it is stated has a good support for personalisation as agents generally are used to represent entities in the real world (e.g. users). Through this approach, the system allows for a completely personal classification of a user's current context. By doing so, it is acknowledged that pre-defined rules does not fit everyone, but should be adapted to give each person a high degree of personalisation. |
