|Thesis abstract: |
In a world made of global interconnections and networking systems, the variety and abundance of available data generates the need for effective and efficient gathering, synthesizing, and querying process, removing information noise. This thesis realizes a system where context awareness is integrated with - yet orthogonal to - data management, where the knowledge of the context in which the data are used drives the process of focusingon currently useful information (represented by means of views), keepinginformation-noise at bay. This activity is called context-aware data tailoring.
The approach proposed in this thesis supports context-aware data tailoring by adopting a powerful context modeling tool known as Context Dimension Tree (CDT), used to define contexts and support the design of accompanying context-dependent views used to assemble contextualized data. The context is evaluated and validated by means of logic in Answer Set Programming (ASP); the discovered combinations of values constituting the current context are verfied and mapped to their corresponding interesting data by the associated views in a uniform process, never leaving the ASP environment.
Inconsistencies have also been taken into account. In fact, sometimes the context information evaluated by the system is compatible with distinct (possibly mutually inconsistent) contexts: in this case, the system has to be ready to provide the user, or the application, with different views, each compatible with one of these contexts and refer the associated contexts for consistency checking. Special cases, in which particular behaviors must be substituted to the standard composition process applying to the context views, are also supported by means of overriding facilities designed to enable a very fine rained control over the contextualization process.
In particular, the proposed approach uses ASP techniques to (i) validate the perceived context against feasible contexts extracted from a CDT provided along with application scenario, (ii) convey to the user the context-dependent views associated to the (possibly multiple) current contexts, (iii) use the views to retain from the underlying dataset only the relevant data for each such context, returning only interesting data and removing noise, (iv) finely control the contexts by means of integrated support for contextual preferences and overriding mechanisms for the views; all this retaining the orthogonality of context modeling to the data while adopting the same framework as for views and data representation.
A prototype is introduced along with experiments and experimental results, considering the actual efficacy wrt. design support, validation and consistency of obtained answers through the tailoring process described in the approach proposed. A real-world application scenario, in which the approach is being actively adopted, is also discussed.