Grid Search Hyperparameters Tuning with Supervised Machine Learning for Awngi Language Named Entity Recognition

Le résumé fourni par la source

Named entity recognition has emerged as a critical step in recognizing, classifying, and extracting the most significant information from unstructured text without human intervention. It is used in information retrieval, conversation systems, machine translation, data mining, and information extraction. However, the Awngi language lacks a NER system due to inadequate datasets. This study combines feature extraction techniques with supervised machine learning algorithms, including support vector machines, Naïve Bayes, and conditional random fields with a grid search algorithm parameter optimization. These supervised machine-learning models often have various hyperparameters that must be tuned to optimize the model's performance by automatically selecting the parameter. Furthermore, the dataset used for training and testing consists of 20,193 annotated Awngi tokens, collected from Amhara Media Corporation, Awngi Telegram, and Facebook pages. The proposed SVM with grid search algorithm approach provided a promising result in ANER with an accuracy of 92% compared with Naïve Bayes, conditional random field, and decision-tree supervised machine learning models. Finally, the experiment results show that the proposed parameter tuning-based model can detect and categorize Awingi-named entities and outperforms the performance accuracy without parameter adjustment.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.