ChatGPT’s performance evaluation for annotating multi-label text in Indonesian language

The high need for artificial intelligence applications in all fields impacts the appropriate dataset for building a good machine. Labeling datasets becomes one of the main tasks needed before training the machine, especially in sentiment analysis. Aspect-based sentiment analysis has more labels in i...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:AIP conference proceedings 2024-05, Vol.3116 (1)
Hauptverfasser: Hakim, M. Faris Al, Prasetiyo, Budi
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The high need for artificial intelligence applications in all fields impacts the appropriate dataset for building a good machine. Labeling datasets becomes one of the main tasks needed before training the machine, especially in sentiment analysis. Aspect-based sentiment analysis has more labels in its process than others. The high number of data also impacts the high cost of processing the data, including the labeling process. Nevertheless, all the problems still need to be solved, including multi-label in Indonesian. It is a potential task that needs to be done by giving several labels in an instance. ChatGPT, as one of the Large Language Models (LLM), has a high potential to carry out the labeling process. ChatGPT-3.5 was examined to label for aspect-based sentiment analysis in the Indonesian language in this study. CASA dataset containing 1080 rows was used to evaluate the performance of the model. The results show that comprehensive exploration must be applied to produce optimal ChatGPT performance for classifying multi-label in Indonesian. The study results will impact the efficiency of the labeling process in multi-label case that needs more effort to be finished.
ISSN:0094-243X
1551-7616
DOI:10.1063/5.0210320