Pay-as-you-go information integration in personal and social dataspaces

A personal and social dataspace is the set of all information pertaining to a given user. It includes a heterogeneous mix of files, folders, email, contacts, music, calendar items, images, among others, distributed among a set of data sources such as filesystems, email servers, network shares, databases, and web servers. In addition, it includes all connections this user has to other users in a number of online services such as social networking web sites. In spite of a personal and social dataspace being richly heterogeneous and distributed, only very limited tools are available to aid users manage their information and have a unified view over it. At one extreme, search engines allow users to pose simple keyword and path searches over all of their data sources. These systems, however, return only ranked lists of besteffort results and provide limited or no means for users to increase the quality of query results returned by the system over time. At the other extreme, systems built on top of classic database technology, such as traditional information-integration systems, provide precise query semantics for queries over a set of data sources. Although the quality of query results returned by these systems is high, they are typically restricted to a subset of the personal information of a user, given the need to specify complex schema mappings to integrate the data. As a consequence, these systems have limited coverage and provide equally limited support for non-expert users to refine their view of their personal information over time. This thesis investigates a new breed of information-integration architecture that stands in-between the two extremes of search engines and traditional information-integration systems. We term this new type of system a Personal Dataspace Management System (PDSMS). Like a search engine, when a PDSMS is bootstrapped, it provides a simple search service over all of the user’s dataspace. In contrast to search engines, however, the PDSMS represents data not at the coarse-grained level of files (or text documents), but rather using a fine-grained graph-based data model. In addition, a PDSMS provides means for a user to increase the level of integration of her dataspace gradually, in a pay-as-you-go fashion. That is done by enabling users to provide simple integration “hints” that allow the PDSMS to improve the quality of query results. In contrast to traditional information-integration systems, however, at no point does the PDSMS require users to specify a global mediated schema for their information. We make four main contributions to the design of PDSMSs. First, we propose the iMeMex Data Model (iDM), a simple, yet powerful, graph-based data model able to represent the heterogeneous data mix found in a personal and social dataspace. Our data model enables query capabilities on top of the user’s dataspace not commonly found in state-of-the-art tools.