Evidence record · UDINUS · SDGs
- Unit
- Rektorat UDINUS
- Period
- 2025
- Goals
- Record
- e33a8264-e289-43c6-a6d4-29c33852193f
- Published
- 2026-10-30
Universitas Dian Nuswantoro (UDINUS), through participating researchers Nurchim, Muljono, Edi Noersasongko, Ahmad Zainul Fanani, and Deshinta Arrova Dewi, conducted a study on the development of a Few-Shot Named Entity Recognition (NER) model for Indonesian sports texts. The activity addressed the challenge of limited annotated resources for Indonesian-language text processing. It examined whether a Few-Shot learning strategy could provide competitive or superior performance while using only minimal labelled examples.
The activity incorporated prototype-based adaptation and prompt-driven fine-tuning into a pre-trained SpaCy pipeline. The proposed model was used to distinguish eight Indonesian sport categories in conditions characterised by severe token imbalance and low lexical diversity. The source describes Indonesian sports headlines as short, context-limited, and structurally compact, and therefore considers Few-Shot methods suitable for capturing fine-grained distinctions among entities in this dataset. The study compared the proposed approach with conventional NER approaches, including SpaCy, BiLSTM-CRF, and transformer-based models, which typically require substantial training data.
The experimental results reported that the Few-Shot model outperformed all baselines, achieving a precision of 0.991 and an F1-score of 0.995. The standard SpaCy model achieved an F1-score of 0.983. The findings indicate enhanced generalization capability, reduced reliance on extensive annotation, and improved discrimination among closely related sport entities. The study identifies Few-Shot NER as a scalable and practically viable approach for low-resource Indonesian natural language processing and domain-specific information extraction where large annotated datasets are unavailable.
Through this activity, UDINUS contributes to research and innovation in computational methods for Indonesian sports text classification. The activity further frames Few-Shot learning as an efficient and adaptable approach for sports analytics, media processing, and broader applications in under-resourced languages. Its institutional relevance is reflected in the development and evaluation of a method intended for domain-specific information extraction under conditions of limited annotated data.