Performance Assessment of Classical Machine Learning Models Used for Entity Relationship Extraction on Online Philippine News Articles

Content from legitimate online news articles is used to determine relationships of objects or entities to enable a more efficient way of extracting information. It could also be used to establish ground truths on news stories or current events to provide a reference against the emergence of fake news articles. This study involved the entity relationship extraction in an online news source, the Philippine Daily Inquirer online site Inquirer.net. Classical machine learning methods were designed to perform pairwise entity relationships via Semantic Web, and the performance of the methods was assessed and compared with one another in terms of training time and F1 scores. A 90–10 train-test split of data was used for each of the methods. A total of 12 machine learning methods were configured to accomplish the task of entity relationship extraction, and a total of 27 top entity relationships were obtained and ranked. The use of support vector machines (SVM) yielded the highest F1 score of 60.73% but took 592.54 seconds to complete the training. The fastest method to train was the multinomial naïve Bayes (MNB) at 0.63 seconds but only obtained an F1 score of 50.14%.