tailieunhanh - Báo cáo khoa học: "Extraction of definitions using grammar-enhanced machine learning"

In this paper we compare different approaches to extract definitions of four types using a combination of a rule-based grammar and machine learning. We collected a Dutch text corpus containing 549 definitions and applied a grammar on it. Machine learning was then applied to improve the results obtained with the grammar. Two machine learning experiments were carried out. In the first experiment, a standard classifier and a classifier designed specifically to deal with imbalanced datasets are compared. The algorithm designed specifically to deal with imbalanced datasets for most types outperforms the standard classifier. . | Extraction of definitions using grammar-enhanced machine learning Eline Westerhout Utrecht University Trans 10 3512 JK Utrecht The Netherlands Abstract In this paper we compare different approaches to extract definitions of four types using a combination of a rule-based grammar and machine learning. We collected a Dutch text corpus containing 549 definitions and applied a grammar on it. Machine learning was then applied to improve the results obtained with the grammar. Two machine learning experiments were carried out. In the first experiment a standard classifier and a classifier designed specifically to deal with imbalanced datasets are compared. The algorithm designed specifically to deal with imbalanced datasets for most types outperforms the standard classifier. In the second experiment we show that classification results improve when information on definition structure is included. 1 Introduction Definition extraction can be relevant in different areas. It is most times used in the domain of question answering to answer What-is -questions. The context in which we apply definition extraction is the automatic creation of glossaries within elearning. This is a new area and provides its own requirements to the task. Glossaries can play an important role within this domain since they support the learner in decoding the learning object he is confronted with and in understanding the central concepts which are being conveyed in the learning material. Different approaches for the detection of definitions can be distinguished. We use a sequential combination of a rule-based approach and machine learning to extract definitions. As a first step a grammar is used and thereafter machine learning techniques are applied to filter the incorrectly extracted data. Our approach has different innovative aspects compared to other research in the area of definition extraction. The first aspect is that we address less common definition patterns also. Second we .