Search-based Automated Resolution For Context Inconsistency

In recent years, with the population of smart devices, context-aware applications are growing quickly and being developed and used widely. These applications can per-ceive environmental status based on contexts derived from various sensors embedded in such smart devices, and offer users desirable adaptive services. For example, the "U-ber" application can use its user’s location contexts to smartly schedule taxi services.However, due to various environmental noises, context inconsistency can occur in many scenarios, which may lead context-aware applications to behave abnormally and cause trouble to their users. It is thus necessary to detect and resolve context in-consistency in time. Although there are quite a few pieces of existing work on context inconsistency detection already delivering satisfactory results, existing work on con-text inconsistency resolution still does not address the problem effectively. The reason is that it is typically difficult to propose a general approach applicable to all scenarios, since environmental noises are usually diversified, complex and scenario-specific. At the same time, we have also found that existing work on context inconsistency reso-lution has largely overlooked two aspects. First, context inconsistency resolution for one constraint can unexpectedly lead to violation of another constraint (i.e., constrain-t interfering problem). Second, context inconsistency resolution may itself affect an application’s functionality or quality in an unexpected way (i.e., side effect problem).Therefore, we in this thesis propose a novel search-based approach to automat-ically resolving detected context inconsistencies for context-aware applications. The approach explicitly takes in account the above two discussed aspects. We try to bal-ance requirements on the resolution’s performance and its effectiveness on alleviating the constraint interfering and side effect problems. We realize this goal by search-ing the whole space by heuristic genetic algorithms, and by reusing previous checking results and avoiding redundant computing. We evaluated our approach experimentally. We compared it with traditional con-text inconsistency approaches under two application scenarios, one of which was sim-ulated based on real-life parameters from field data, and the other ran with realistic taxi data from a city-wide application. The experimental results reported that our ap-proach is effective in addressing the constraint interfering and side effect problems, while resolving detected context inconsistencies effectively and efficiently.We have also implemented supporting components and demo systems for our ap-proach. First, we implemented some context management components, which can assist software developers in their tasks of detecting and resolving context inconsisten-cies for their context-aware applications. Second, we implemented some demo systems for illustrating and comparing the effectiveness of different context inconsistency res-olution approaches under different platforms (e.g., PCs and Android devices).