Sri Hadianti, Dwiza Riana, Herdian Tohir, Jarwadi Jarwadi, Tjaturningsih Rosdiana, Evi Sopandi, Dinar Ajeng Kristiyanti

WORD2VEC OPTIMALIZATION USING TRANSFER LEARNING IN INDONESIAN LANGUAGE FOR HIGHER EDUCATION

Introduction

Word2vec optimalization using transfer learning in indonesian language for higher education. Optimize Word2Vec for Indonesian higher education using transfer learning & lexicon expansion. NewWord2Vec model improves word detection by 4%, capturing linguistic nuances for NLP applications.

Abstract

Natural language processing (NLP) in Indonesian faces challenges due to limited linguistic resources, particularly in developing optimal word embedding models. This study optimizes the Word2Vec model for Indonesian in higher education contexts by leveraging transfer learning and lexicon expansion. Using a dataset of 4,463 higher education related tweets consisting of positive and negative sentiment categories, the proposed NewWord2Vec model combined with a Support Vector Machine (SVM) classifier achieved a 4% improvement in word detection accuracy compared to the standard Word2Vec. This enhancement demonstrates better performance in capturing linguistic nuances and sentiment orientation in Indonesian text. However, the model’s applicability remains limited to higher education terminology, and potential biases from transfer learning must be addressed. Future research should expand the dataset to diverse domains and refine the transfer learning process to better capture contextual variations in Indonesian. These findings contribute to advancing NLP applications in Indonesian, particularly for automated assessment systems, recommendation tools, and academic decision-making processes

Review

This study tackles a pertinent challenge in Indonesian Natural Language Processing (NLP): the optimization of word embedding models, specifically Word2Vec, in a resource-constrained environment. By leveraging transfer learning and lexicon expansion, the authors propose a "NewWord2Vec" model aimed at improving word detection accuracy within higher education contexts. The abstract suggests a commendable effort to address the scarcity of linguistic resources for Indonesian, a critical step towards developing more robust NLP applications for the language. The reported 4% improvement in word detection accuracy using the proposed model combined with an SVM classifier is presented as a significant advancement, indicating the potential of their approach. A key strength of this work lies in its specific focus and demonstrated performance gain. The targeted optimization for higher education terminology in Indonesian, a domain with distinct linguistic characteristics, is a valuable contribution. The combination of transfer learning and lexicon expansion to enrich the Word2Vec model is an innovative strategy to overcome data limitations and enhance the model's ability to capture nuanced meaning and sentiment in a challenging language. This advancement holds promising practical implications for various academic applications, such as automated assessment systems, content recommendation tools, and supporting academic decision-making processes, as highlighted by the authors. Despite the positive results, the abstract also transparently outlines several limitations that warrant consideration. The model's applicability is explicitly noted to be restricted to higher education terminology, which naturally limits its generalizability across broader Indonesian domains. Furthermore, the potential biases introduced through transfer learning are acknowledged as an area requiring further investigation, which is crucial for ensuring fairness and accuracy in downstream applications. While a 4% improvement is reported, a deeper understanding of the experimental setup, baseline models used for comparison (beyond "standard Word2Vec"), and a detailed error analysis would strengthen the overall claims. Future work should indeed prioritize expanding the dataset and refining the transfer learning process to enhance contextual understanding and mitigate biases, paving the way for more broadly applicable and robust Indonesian NLP solutions.

Full Text

Comments

You need to be logged in to post a comment.