Management of unspecified semi-structured data in multi-agent environment

Amounts of available heterogeneous semi-structured data grow rapidly on the Web and other data repositories. This raises the need to provide simple and universal ways to access this data. To provide such an interface, we propose to exploit the notion of "unspecified ontologies", describing the data objects as a list of attributes and their respective values. In order to facilitate an efficient management of the unspecified data objects we use a multi-agent channeled multicast communication platform. The data objects are stored distributively, such that each attribute is assigned a designated channel. This allows performing efficient searches by parallel querying of the relevant channels only, and aggregating the partial results. Moreover, the multi-agent platform facilitates advanced data management through extracting metadata from the data objects. We implemented a prototype system and experimented with a corpus of real-life E-Commerce advertisements. Our results demonstrate scalability of the proposed approach and the accuracy of the extracted meta-data.