File Training Generator For Indonesian Language In Named Entity Recognition Using Anago Library

Abstract

Named Entity Recognition (NER) or Named Entity Recognition and Classification (NERC) is one of the main components of an information extraction task that aims to detect and categorize named entities in a text. NER is generally used to detect people's names, place names, and organization of a document, but can also be extended to identify genes, proteins, and others as needed. NER is useful in many NLP (Natural Language Processing) applications such as question-answering, summaries, and dialog systems because it can reduce ambiguity. NER also deals with other information extraction tasks such as relation detection, event detection, and temporal analysis. To avoid this need to train data source. The data train can be taken from various sources of news/articles crawled on the internet. The news will then be annotated by users with various labels. The news/article sources are in the thousands, while to make this training by using file is manual. And sometimes there is an error because this manual was made when it will form the NER model as needed. This research will be made so that training files can be assisted by using applications so that the error rate can be smaller or there will be no errors.

Extract antibody and antigen names from biomedical literature

“…Developed and optimized by Nakayama in 2017 with the combined technique BiLSTM-CRF [26], anaGo was implemented in Keras for NER and many other sequence labeling tasks. anaGo implements different pre-trained word embeddings as input; it also has the capability to self-generate word embedding based on training data [12,27]. The BiLSTM-CRF architecture is described in Fig.…”

“…Bi-LSTM uses two LSTM networks, forward (f1-4) and backward (b1-4). The vector representations from both networks are concatenated (c1-4) and inputted to the CRF tagging layer for label assignment [12,13,27]. The model consists of 10 layers with over 2 M parameters.…”