In the last years, research on adaptive Information Extraction from text (IE) has largely focused on algorithms and systems adaptable to new Web-related applications/scenarios by users with analyst's knowledge, i.e. knowledge on the domain/scenario only, [Kushmerick 1997], [Califf 1998], [Freitag 1999], [Soderland 1999], [Freitag 2000], [Muslea 2000], [Ciravegna 2001]. Successful commercial products have been created and there is an increasing interest on IE in the Internet market. The more the focus is on the user, the more the need for user-specific tools arises. Most of the current approaches are based on an adaptation phase in which the user provides a set of texts with the relevant information highlighted or with associated filled templates. Tagging is just one part of the adaptation task, though, in building real world applications. Adaptation as a one-way process (from tagged examples to rules) is unlikely to provide optimized results for specific users, as different uses will require different types of results (e.g., high recall in some cases, high precision in others). There is the necessity, we believe, to fully support users during the whole adaptation process so to maximize effectiveness and appropriateness of the final application, an to minimize the burden of system adaptation. In this paper we discuss requirements for user involvement in application development in Amilcare, a system for adaptive IE
User Involvement in Adaptive Information Extraction: Position Paper
2001-01-01
Abstract
In the last years, research on adaptive Information Extraction from text (IE) has largely focused on algorithms and systems adaptable to new Web-related applications/scenarios by users with analyst's knowledge, i.e. knowledge on the domain/scenario only, [Kushmerick 1997], [Califf 1998], [Freitag 1999], [Soderland 1999], [Freitag 2000], [Muslea 2000], [Ciravegna 2001]. Successful commercial products have been created and there is an increasing interest on IE in the Internet market. The more the focus is on the user, the more the need for user-specific tools arises. Most of the current approaches are based on an adaptation phase in which the user provides a set of texts with the relevant information highlighted or with associated filled templates. Tagging is just one part of the adaptation task, though, in building real world applications. Adaptation as a one-way process (from tagged examples to rules) is unlikely to provide optimized results for specific users, as different uses will require different types of results (e.g., high recall in some cases, high precision in others). There is the necessity, we believe, to fully support users during the whole adaptation process so to maximize effectiveness and appropriateness of the final application, an to minimize the burden of system adaptation. In this paper we discuss requirements for user involvement in application development in Amilcare, a system for adaptive IEI documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.