Design of Named Entity Recognition Model for Awingi Language Using Deep Learning Approach

Abstract

Named entity recognition is a field of NLP that aims to detect and extract relevant information from unstructured text documents. However, some Ethiopian languages were getting started in different research fields. One of the languages spoken in Ethiopia is called Awingi language has a rich morphology and it is highly lacking in computational linguistic tools. There is some NER research done in Ethiopian language using different approaches and classification techniques, but since there was a lack of study conducted on NER for the Awingi language. This study aims to design a named entity recognition model for the Awngi Language using a deep learning approach with word2vect word embedding techniques. The data collected from AMICO Hiber Awiꬼgi program was posted on social media pages, Awngi Language and Literature Department of Injibara University, and Awi Zone Educational and Training Department. Therefore, the study uses a newly tagged dataset with 37,133 tokens. Presently, researchers focused on discovering four main named entities such as persons, locations, Dates, and organizations from unstructured Awiꬼgi text by using the three algorithms namely convolutional neural networks, long short-term memory, and bi-directional long-short-term memory. For the feature engineering task, Word2vec and one-hot encoding are utilized. To identify the best-performing model, the researchers have conducted several experiments. Over other algorithms, the Bi-LSTM algorithm before sampling and word embedding performed better results, i.e 95.4%, 92.4%, 95.3%, and 91.2% Accuracy, precision, recall, and f1-score, respectively, and after sampling and word embedding it performed with 99.4%, 97.3%, 99.2%, and 97.2% Accuracy, precision, recall, and f1-score, respectively.