Istituto di Scienza e Tecnologie
dell’Informazione “A. Faedo”

The Laboratory is mainly concerned with:

1. Automatic Text Classification. The research applies machine learning techniques for automatic categorization of textual documents. This approach is very useful to organize large document collections in semantically meaningful categories, when a suitable training set of documents is supplied for each category.
2. Knowledge Representation, Reasoning and the Semantic Web. The research addresses logics (in particular, logic programming and description logics) for KRR and the Semantic Web, the management of uncertainty, top-k retrieval, and logic-based approaches to (distributed) multimedia information retrieval and ontology.

The research combines database and information retrieval techniques to support new classes of applications of increasing importance, such as multimedia digital libraries and multimedia object repositories (e.g., audio-visual or image content, Learning Object digital libraries), multimedia documents in the Web. The unifying concept in these applications is feature-based similarity search.

The concept of similarity searching based on relative distances between a query and database objects has become essential for a number of application area, e.g. data mining, signal processing, geographic databases, information retrieval, or computational biology. The growing need to deal with large, possibly distributed, archives requires an indexing support to speedup retrieval. The research focuses on the study and experimental evaluation of new data structures supporting the similarity retrieval of multimedia information. Multimedia information is usually represented through a set of features, representing the characteristics of the objects, such as: words for text documents; color, texture, and shapes for images; moving objects, key frames, and camera operations for video. Similarity based access means that the user specifies some characteristics of the wanted information objects, and the system retrieves the most relevant objects with respect to the given characteristics, i.e., the objects most similar to the query. Such approach assumes the ability to measure the distance (with some kind of metric) between the query and the database objects, and, obviously, between the database objects. In particular, we study new techniques improving the efficiency of retrieval when features are organized in multidimensional indexes and generic metric spaces. Moreover, we are working on the definition of conceptual structures able to bridge so called "semantic gap", i.e., the gap between low level extracted features at and the real information content. Since we use XML to describe the (partial) structure of multimedia complex objects, and XML is also the description language for the standards on multimedia object content (e.g., MPEG7), we have developed searching techniques to efficiently support similarity access to both multimedia and XML data. As a result of the research activities, in the context of several research projects, we have developed our own Multimedia Content Management System, called MILOS, that allows the implementation of specific applications for searching, enhancing and delivery multimedia content. This system has been proven successful in building specific applications managing different multimedia content, such as historical films ( http://milos.isti.cnr.it:5900/milos/echo.html), personal and shared photos (http://milos.isti.cnr.it/milos/album/), scientific literature ( http://milos.isti.cnr.it:5900/milos/sigmodDblp.html), etc.

Our research in this field addresses three important issues:

1. the definition of a reference model and a reference architecture for digital libraries
2. the design and implementation of large scale digital libraries infrastructures
3. the design and implementation of large scale digital repositories infrastructures.

All these research activities are carried out in the context of large EU funded projects. In particular, our team has embarked on an effort to formalize a definition of a Digital Library system: a conceptual framework describing the architectural, functional, operational and behavioral characteristics of this particular kind of information system. The outcome of this effort should contribute significantly to the understanding of digital library technology and should serve as a yardstick of quality and richness both with respect to achievements in the field so far and for the generic digital library management systems that will be developed in the future, since it will specify their expected features and properties.

A fundamental issue in digital libraries is their underlying architecture which must be capable of enabling all desirable performance characteristics, i.e., large scale information distribution, scalability, availability, robustness, reliability, self-organization, adaptability etc. Recent advances point toward three architectural alternatives: peer-to-peer architectures, grid middleware and service-oriented architectures.

1. Peer-to-peer architectures allow for loosely coupled integration of information services and sharing of information such as recommendations and annotations.
2. Grid middleware is appropriate because future digital libraries will integrate, in a secure way, information that is broadly distributed. Furthermore, certain services within digital libraries are computationally intensive and complex (e.g., extraction of features in multimedia documents to support content-based similarity search).
3. Finally, service oriented architectures provide mechanisms to describe the semantics and usage of information provision functionality. They also support mechanisms to combine services into workflow processes for sophisticated search and maintenance.

Our team has developed prototypes implementing all three of these architectural directions and has evaluated experimentally their impact on digital library system architectures. Digital repositories are, in some sense, a special case of digital libraries. Our team is now studying the common ground between the two areas, focusing especially on issues of access policy (mostly open access), operational environment and the creation and management of digital repositories. An additional focus is on the development of an infrastructure of repositories to support research and learning.