International Journal on Advanced Science, Engineering and Information Technology, Vol. 10 (2020) No. 3, pages: 1123-1129, DOI:10.18517/ijaseit.10.3.10939

An Evaluation Methodology of Named Entities Recognition in Spanish Language: ECU 911 Case Study

Marcos Orellana, Andrea Trujillo, Juan-Fernando Lima, María-Inés Acosta, Mario Peña

Abstract

The importance of the gathered information in Integrated Security Services as ECU911 in Ecuador is evidenced in terms of its quality and availability in order to perform decision-making tasks. It is a priority to avoid the loss of relevant information such as event address, places references, names, etc. In this context it is present Named Entity Recognition (NER) analysis for discovering information into informal texts. Unlike structured corpus and labeled for NER analysis like CONLL2002 or ANCORA, informal texts generated from emergency call dialogues have a very wide linguistic variety; in addition, there is a strong tending to lose important information in their processing. A relevant aspect to considerate is the identification of texts that denotes entities such as the physical address where emergency events occurred. This study aims to extract the locations in which an emergency event has been issued. A set of experiments was performed with NER models based on Convolutional Neural Network (CNN). The performance of models was evaluated according to parameters such as training dataset size, dropout rate, location dictionary, and denoting location. An experimentation methodology was proposed, with it follows the next steps: i) Data preprocessing, ii) Dataset labeling, iii) Model structuring, and iv) Model evaluating. Results revealed that the performance of a model improves when having more training data, an adequate dropout rate to control overfitting problems, and a combination of a dictionary of locations and replacing words denoting entities.

Keywords:

named entity recognition; Spanish language; emergency calls; informal text.

Viewed: 133 times (since Sept 4, 2017)

cite this paper     download