| |
The system to be developed has the following main objectives that will be achieved by the innovative techniques sketched below, for each main objectives:
- to define a modular system architecture, constituted of three main components, described in the following three subsequent points. The results of the system requirement analysis and the definition of the system architecture will constitute the milestone 1 and 2 respectively (ML1 and ML2)
- to define a system component which performs a first filtering phase, which is aimed at rapidly proposing at the user the information selected based on his/her time-dependent personal profile ( push phase ).
- to define a system component which performs a pull phase on the basis of the user’s indications. In this phase a query is automatically constructed and submitted to a distributed multimedia information retrieval system gathering additional information from other archives, from the web and from a local repository.
- at this point the retrieved information (both those resulting from the first phase and those resulting from the second phase) has to be edited and prepared for presentation. This is a quite new phase with respect to the usual electronic news composition systems. It requires the application of techniques of multi-document and multi-media visualisation and summarisation that take into account the trust scores (and can be viewed as the definition of a basic "electronic writer").
- the integrated system will produce “electronically composed” pieces of news and will present them to the user in a personalised way with new modality of interaction and information presentation so that the system can be used on different platforms to suit mobile users.
- At any point in this process the user can store in a local repository any piece of news he thinks might come useful later. This enable journalist and editors to build up a personal archive of news they consider relevant and useful for their profession.
 |
 |
This process is sketched in Figure 1
A system based on such a process would be particularly useful for content programming of entertainment news, since salient characteristics of this kind of news as that they are highly heterogeneous, multimedia, and originated from a large variety of source with different level of trust associated to them.
Strategic impact ::
The availability of a dynamic and personalised composition of the multimedia information content provided by a plurality of information sources is of strategic importance to Europe. It allows contributing to develop and reinforce personal and original points of views on topical questions and more in general to build a mature democracy expressing a melting-pot culture. Moreover it is one of the critical technology areas participating to the general content industry issues. PENG should become the compromise to be traded-off between the computer industry (US dominated) and the European consumer electronics, broadcast and broadband industry.
Innovation-related activities ::
PENG can greatly contribute to innovate the way of generating information, for the benefit of the journalist themselves and for the general users of news services (e.g. TVs, radios, newspapers, Web, etc.). Professionals, such as journalists or editors, can tune the contribution of the distinct sources to their information gathering, filtering and editing tasks, by specifying queries expressing constraints on the multimedia and time-dependent content of the news so as to focus on a particular event, and by associating distinct trust scores with the information sources.
From a technical point of view the main innovative characteristic of such a system are:
- to "tune" the contribution of the distinct content types and sources of information (we consider several distinct sources of information from which to extract the news).
- to apply soft computing techniques for modelling both news content and categorisation and the user context. In particular, the central issues to this modelling activity will be:
- Dealing with uncertainty and imprecision in the user-system interaction: human communication is often ambiguous, something rarely noticed in inter-human communication (use of probability theory, fuzzy logic).
- Dealing with vagueness and imprecision in the filtering task when associating news to a category subject matter or topic area: news can belong to several categories and deal at the same time with different subject matters to a different strength and deepness (use of categorization techniques and fuzzy set theory).
- Learning the user context: disambiguation using pre-learned (global or personal) knowledge or dynamically learned user preferences changing in time allows a more natural expression of the user's need and makes the system adaptive to the user context (use of content based and collaborative filtering techniques).
- Representing users' interests through ontologies: in order to capture, learn and reason with meaningful information, the definition of concept-based representations which meaningfully express the users’ interest is of extreme importance. To this aim, currently defined ontologies in the news domain will be employed so that the systems developed will adhere to current standards.
- To define methods of document visualisation and summarisation that take into account the relevance score, the trust score and the similarities between the content of news from different sources on the same topic. In this context multi-document summarisation and visualisation techniques will be employed.
- to define and apply flexible methods of information categorisation
- to explore multi-modal presentation, i.e. the possibility that the developed system can be used on different platforms (desktop computers, mobiles, palmtop computers, etc.), changing the presentation of the information to suit the different characteristics of the platform or device used to access it. This will enable access to the service anytime and anywhere for user (especially journalists) that are extremely mobile.
The system will be conceived as an open source, distributed system, based on a client-server architecture all the personalisation tasks will run on the client (i.e., the news categorisation and user profile generation modules, the trust scores module customisation and summary production) while the retrieval task will exploit the customised information provided by the client to retrieve the news that will be pulled back to the client for the final presentation to the user.
|
|
objectives & innovation