Portuguese Named Entity Recognition using Conditional Random Fields and Local Grammars

Portuguese Named Entity Recognition using Conditional Random Fields and Local Grammars Juliana P. C. Pirovani, Elias de Oliveira Universidade Federal do Esp´ırito Santo - UFES Av. Fernando Ferrari, 514, 29075-910 Vit´oria, ES, Brazil juliana.campos@ufes.br,elias@lcad.inf.ufes.br Abstract Named Entity Recognition involves automatically identifying and classifying entities such as persons, places, and organizations, and it is a very important task in Information Extraction. Conditional Random Fields is a probabilistic method for structured prediction, which can be used in this task. This paper presents the use of Conditional Random Fields for Named Entity Recognition in Portuguese texts considering the term classification obtained by a Local Grammar as an additional informed feature. Local grammars are handmade rules to identify named entities within the text. The Golden Collection of the First and Second HAREM considered as a reference for Named Entity Recognition systems in Portuguese were used as training and test sets respectively. The results obtained outperform the results of competitive systems reported in the literature. 1. Introduction Named Entity Recognition (NER) aims at automatically identifying and classifying entities such as persons, places, organizations and values. This is a fundamental task in In- formation Extraction since, besides having several applica- tions, other tasks such as relations and events extraction, question answering systems and entity-oriented search de- pend on it as a preprocessing step (Jiang, 2012). NER is not a simple task. Several categories of named en- tities (NEs) are written similarly and they appear in similar contexts. In addition, the same NE can be classified into different categories depending on the surrounding context and some entities do not appear even in large training sets. Thus, dictionaries are not always useful. In 1995, the Message Understanding Conference (MUC-6) (Grishman and Sundheim, 1996) included the NER task for the first time for the English. Several similar events have emerged later such as the CoNLL (Tjong Kim Sang and De Meulder, 2003) and the HAREM (Santos and Cardoso, 2007; Mota and Santos, 2008). HAREM was an initiative for the Portuguese and the annotated corpora used in the First and Second HAREM, known as the Golden Collec- tions (GC), are used as a golden standard reference for NER systems in Portuguese. HAREM differs from other similar events in two aspects (Mota and Santos, 2008): the classification of an NE de- pends only on its use in context and more than one classi- fication can be assigned to an NE. Moreover, the HAREM classifies 10 categories of NEs (Person, Place, Organiza- tion, Value, Time, Event, Abstraction, Work, Thing and Other). Thus, the HAREM presents a more demanding task and, therefore, the performance values obtained using its reference data sets are still lower compared to the others (Santos and Cardoso, 2007). NER systems can be developed using the following ap- proaches: linguistics, machine learning or hybrid. This work seeks to explore the potential of the linguistics and machine learning approaches by constructing a hybrid system for NER in Portuguese. The presented strategy, CRF+LG, combines a labeling obtained by a Conditional Random Fields …

Content truncated to 500 words