A Sequence Model Approach to Relation Extraction in Portuguese

A Sequence Model Approach to Relation Extraction in Portuguese Sandra Collovini, Gabriel Machado, Renata Vieira PUCRS University – Porto Alegre – Brazil sandra.abreu@acad.pucrs.br, gabriel.machado.002@acad.pucrs.br, renata.vieira@pucrs.br Abstract The task of Relation Extraction from texts is one of the main challenges in the area of Information Extraction, considering the required linguistic knowledge and the sophistication of the language processing techniques employed. This task aims at identifying and classifying semantic relations that occur between entities recognized in a given text. In this paper, we evaluated a Conditional Random Fields classifier for the extraction of any relation descriptor occurring between named entities (Organisation, Person and Place categories), as well as pre-defined relation types between these entities in Portuguese texts. Keywords: Information Extraction, Relation Extraction, Conditional Random Fields 1. Introduction Information Extraction (IE) is a process of getting structured data from unstructured information in the text (Bach and Badaskar, 2007; Jurafsky and Martin, 2009). After this structured information can be used by a wide range of NLP applications. Usually, IE can be regarded as a pipeline process, in which some type of information is extracted at each step. Relation Extraction (RE) is one of the stages of IE, which aims to identify and classify semantic rela- tions that occur between entities recognized in a given text (Bach and Badaskar, 2007; Jurafsky and Martin, 2009). The two major types of RE are closed domain and open domain (Banko and Etzioni, 2008): closed- domain RE systems consider only a closed set of rela- tions between two arguments, while open-domain RE systems do not need a pre-specified definition of the relation. Currently, there are plenty of systems for RE from unstructured data and there are different methods for dealing with this task. Among supervised methods stands Conditional Random Fields (CRF), which are very powerful for segmenting and labeling sequential data (Lafferty et al., 2001). CRF have now become almost a standard for the task of Named Entity Recog- nition (NER) (McCallum and Li, 2003), and have more recently been applied to the task of RE from text (Cu- lotta et al., 2006; Banko and Etzioni, 2008; Wu and Weld, 2010; Li et al., 2011; Collovini et al., 2014). In this paper, we evaluated a CRF classifier in two scenarios: the extraction of any relation descriptor occurring between named entities (Organisation, Per- son and Place categories), and the extraction of rela- tion descriptors expressing pre-defined types of rela- tions between these entities (“affiliation” and “place- ment” relations). We define a relation descriptor as the text chunks that describe the explicit relation, occur- ring between a pair of named entities in the sentence. For example, we have the relation descriptor “pro- fessor at” between the Person named entity “Hugo Dom´ech” and the Organisation named entity “Univer- sidade Jaume de Castell´on” in the sentence: “Hugo Dom´ech, professor da Universidade Jaume de Castell´on.” (Hugo Dom´ech, professor at Universidade Jaume de Castell´on.) This work is organized as follows. In Section 2, we review the related work. The RE model is described in Section …

Content truncated to 500 words