CORDIS bietet Links zu öffentlichen Ergebnissen und Veröffentlichungen von HORIZONT-Projekten.
Links zu Ergebnissen und Veröffentlichungen von RP7-Projekten sowie Links zu einigen Typen spezifischer Ergebnisse wie Datensätzen und Software werden dynamisch von OpenAIRE abgerufen.
Leistungen
Consorzio Pisa Ricerche (CPR) is a non-profit organisation set up to promote and co-ordinate the transfer of innovative technology and expertise from the university and research environment to the industry and other areas of application. CPR - META Centre has developed during BIZON project a Knowledge Management Description Language, which is also usable outside the BIZON project:
- KMDL definition: A language has been defined by CPR META KMDL (Knowledge Management Description Language) configure & customise the system and make it homogeneous for diverse industrial realities.
- Graphic Interface for KMDL: In order to set the parameters described previously without resorting to low-level editing of obscure configuration files, we have designed a graphical tool to enable the administrator set up configuration parameters in a comfortable way. The backend for this tool is of course KMDL, whose syntax is interpreted by the tool and used to store the administrator's preferences. KMDL syntax, which extends the XML taxonomy definition, is loaded at startup, so that the tool loads in one take both the taxonomy structure and the current configuration.
OntoBroker is the leading ontology-based inference engine. Its performance is the basis for intelligent solutions, which perform on a semantic middleware, delivering explicit and implicit information form various data sources and applications. OntoBroker is highly scalable through distribution and integration of commercial database systems.
At the heart of the Semantic Web lies the meaning of concepts, contexts and relations. The inference engine OntoBroker® processes ontologies. Our system excels by background knowledge from these models. Thus, new knowledge is derived.
Ontologies serve as a means for establishing a conceptually concise basis for communicating knowledge for many purposes. OntoBroker® allows the intelligent processing of ontologies. Thereby, the overall goal is the exploitation of rich semantic structures for machine-supported access to explicit and implicit knowledge. This purpose drives our development of OntoBroker® and our investigations of the Semantic Web applications of the future.
BIZON has added valuable improvements to the above, including the development of technology for integrating machine learning and ontologies for means such as personalization and anticipation. The management of the ontologies has been extended in order to support the collaboration of multiple users as well as to support the management of multiple ontologies in parallel.
Furthermore, BIZON has also provided the possibilities for implementing and testing this technology in content management and corporate portal scenarios.
BIZON intelligent Customer Relationship Management (iCRM) provides a model on how to improve the way dispersed, incomplete and heterogeneous customer related data are transferred into relevant business information by combining ontologies and machine learning techniques. The BIZON model shows through a combination of search, personalisation and anticipation functionalities, how intelligent Customer Relationship Management can be implemented without creating the need to change all existing business processes in the company in order to fit an overall customer data model.
The BIZON enhanced Corporate Portal pilot is based on a Ksolutions technological platform that allows handling all the corporate knowledge independently of the structure of data. The Ksolutions CP represents a solution that provides a single point of access to all the information present in the company and is also able to optimise all those business processes of the company, in particular it allows an easy retrieval of information, thus improving the productivity and the quality of decisions. Users accessing the portal are, generally, interested in different kinds of services and require to access to a portal service in a different way; in order to answer to these requirements, document personalized searching and anticipation services can improve the quality and the effectiveness of their activities on the CP.
Within BIZON it was enhanced with the following features:
- Content categorization: as from the first pilot
- Multiple and dynamic user profiling: A profile is the dynamic (and automatic) representation of the user behaviour. User behaviour and interests are dynamic, as they change in time, and context-dependent as users are involved in different activities at the same time; a user is associated with multiple profiles, each one appropriated for different user contexts. All the information needed by a profile is automatically learnt by the system using the set of documents the user has considered relevant in different contexts.
- Personalisation: Upon a searching request/session. It is important to restrict the number of documents returned in searches, typically too large, to the ones really relevant for the user., the corporate portal returns a list of documents sorted and highlighted depending on the relevance degree based on the current user profile.
- Anticipation. Document anticipation is another form of personalisation, where the system proactively presents new documents that the system judges interesting for the user.
The Content Management and Delivery System Pilot of BIZON is based on a content/service delivery system developed and managed by Ksolutions, where the process of content acquisition/production, management and delivery represents the central part of the value-chain. The main pilot objective was to integrate Content Management and Delivery System (CMDS) with a common (shared) Knowledge base (Ontology based), to allow the automatic classification of documents and to ease the searching and browsing of information.
The Content Management System (CMS) is responsible for providing a set of document management functionalities by which documents of any format can be stored and managed. For each document it is possible to add metadata that ease the searching and browsing of information, specially for documents without content (as is the case for images, e.g. scanned invoices or the likes). The enhanced CMS is compliant with the Dublin Core Metadata Initiative (DCMI).
Examples of metadata are:
- Abstract: an explicit description of the key concepts included in the documents, as defined by the editors when documents are stored in the system.
- Taxonomies and categories associated with the document: again this association should be done explicitly by the contents editor when documents are stored in the system.
- Author, creation date, etc.
These metadata represent the structured pieces of information that it is normally possible to associate with any document in the CMS. Both documents and their explicit metadata are the starting point for knowledge management techniques in order to extract and infer other concepts that can ultimately enrich the metadata set.
Usually, document annotation (i.e. adding metadata) is manually done; within BIZON, a semi-automatic process was developed to assist the user when annotating new documents to be inserted into the CMS for the automatic classification of documents into a set of predefined taxonomies.
The modules that Consorzio Pisa Ricerche - META Centre has developed during BIZON project are the following:
- Text Analysis Module: this module gives a document representation suitable to the application of ML techniques. It includes:
-- Tokenisation;
-- Stemming (for different languages);
-- Stop-words removal (for different languages);
-- Vocabulary based representation.
- CAT Module: Categorisation is the main objective of the categorisation module. The target of categorisation is the taxonomy imposed by the system, while relevant features are extracted from annotated documents. Categorisation is a typical task of supervised learning. In BIZON, categorisation is:
-- Hierarchical, since we have full taxonomies;
-- Based on both document contents and metadata.
The taxonomy to classify the document onto is not necessarily flat. Hierarchical categorisation can be obtained by recursive categorisation. There is one binary classifier per node in the taxonomy. Whenever a classifier determines that the document belongs to the category, classification continues recursively on the node children, until a leaf or a negative categorisation is reached. For each node, only one (possibly different from all other nodes) feature vector is constructed to be used by the children to classify the document. This means that the classifiers of the children of a node all classify basing on the same feature vector. It is obtained by text processing and/or queries over the ontologies. Text processing is performed using Text Analysis Module and ontology queries are performed by means of Ontobroker.
- User Profiling module: this module implements the user profiling used in the WP6 to carry out personalisation and anticipation. This code integrates and reuses much of the code already developed in the CAT module and in the Text Analysis Module.
- SVM module: the Machine Learning Algorithm used in the system is the SVM algorithm developed by CPR-META.
Suche nach OpenAIRE-Daten ...
Bei der Suche nach OpenAIRE-Daten ist ein Fehler aufgetreten
Es liegen keine Ergebnisse vor