Salsa Dillah, Lutfiah Tri Syahyaningsih, Dewi Fatmarani Surianto, Nur Azizah Eka Budiarti, Dyah Darma Andayani
The measurement of semantic similarity between sentences in Indonesian still faces challenges, especially in capturing contextual meanings that cannot be represented by rule-based approaches such as the Kamus Besar Bahasa Indonesia (KBBI). This study aims to evaluate and enhance the semantic representation of Indonesian sentences using embedding-based representation learning methods. Five models were compared: TF-IDF, FastText, IndoBERT, pre-trained SBERT, and the proposed method SBERT fine-tuned using the triplet loss approach. The dataset consists of 3,000 sentence pairs constructed from 500 KBBI entries, each labeled as a synonym or non-synonym. The evaluation results show that the proposed model achieves the highest accuracy of 92% and an F1-score of 0.92, followed by IndoBERT (90%), pre-trained SBERT (82%), FastText (65%), and TF-IDF (51%). The integration of the triplet loss effectively optimizes the vector positioning of sentence representations in the semantic space, allowing sentences with similar meanings to be closer and those with different meanings to be further apart. These findings demonstrate that the proposed approach is capable of capturing complex semantic nuances and contextual variations in Indonesian, contributing significantly to the development of more accurate meaning-based models for various natural language processing tasks, such as semantic similarity measurement and text classification. © The Korean Institute of Intelligent Systems. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/) which permits unrestricted non commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Department of Informatics and Computer Engineering, Universitas Negeri Makassar, Makassar, Indonesia
