Objectives of the project

O1. Development of new retrieval techniques, based on probabilistic models, for structured documents in XML taking into account context information, both from the user (focussing the search only to the relevant material) and the documents themselves (determining the most appropriate granularity of the results).

O2. Design of novel search personalization techniques, capable to model, extract, and handle different kinds of contextual information (such as the dynamic ones related to ongoing retrieval tasks, or the static ones that can be obtained from domain knowledge), and use them for a better function of long-term system knowledge about the user, resulting in more accurate personalized actions. One key research topic is the data-mining of the information obtained from the interaction of the user with the system, particularly the exploitation of the query -logs.

O3. Define new models of redundancy (i.e. models that effectively combine estimations of relevance with estimations of redundancy) apply them to post-process retrieval results for text retrieval, web retrieval, xml retrieval and summarisation which is a wide task among all the objectives of this project: multidocument summarisation (clusters), personalised summaries, document representation, etc.

O4. Define new models of subtopic diversity and use them to enhance the coverage of subtopics in retrieved set of documents. Additionally, study how to use these models in the context of passage retrieval and automatic summarization.

O5. Design of new probabilistic classification methods for plain and structured documents in XML, using hierarchies of classes (i.e. based on thesauri), or without any organization. Also the design of new clustering methods able to deal efficiently with large collections of different media sources: xml, text, web, video, etc.

O6. Research of enhanced knowledge and context representation techniques aimed to tackle key problems such as the user information sparsity inherent to large-scale retrieval spaces. Development of new recommendation techniques under this framework.

O7. Research of dynamic multi-criteria relevance aggregation models for the combination of the multiple retrieval dimensions developed in the project into aggregate, unified views.

O8. Application of previous and new results in terms of models, techniques and efficient implementations to the same tasks: analysis, indexing, retrieval, browsing, visualisation, summarisation, clustering and classification, to different media sources beyond text: video, audio and web-news.

Project RIM3 -  Webmaster -