WIBOWO, ALBERT CHRISTIAN HYPERPARAMETER ANALYSIS FOR FINE-TUNING BERT-BASE-CASED IN CURRICULUM VITAE INFORMATION EXTRACTION. Project Report. UNIVERSITAS KATOLIK SOEGIJAPRANATA, Semarang.
|
Text
22.K1.0008-ALBERT CHRISTIAN WIBOWO-COVER_a.pdf Download (831kB) | Preview |
|
![[img]](http://repository.unika.ac.id/style/images/fileicons/text.png) |
Text
22.K1.0008-ALBERT CHRISTIAN WIBOWO-ISI _a.pdf Restricted to Registered users only Download (1MB) |
|
|
Text
22.K1.0008-ALBERT CHRISTIAN WIBOWO-DAPUS _a.pdf Download (736kB) | Preview |
|
|
Text
22.K1.0008-ALBERT CHRISTIAN WIBOWO-LAMP _a.pdf Restricted to Registered users only Download (933kB) |
Abstract
This research focuses on analyzing hyperparameter tuning in the fine-tuning process of a BERT-based model for information extraction from Curriculum Vitae (CV) datasets using Natural Language Processing (NLP), specifically Named Entity Recognition (NER). The main objective of this study is to investigate the influence of several hyperparameters, including batch size, learning rate, number of training epochs, and maximum sequence length, on the performance of the BERT model in extracting information from unstructured CV text. The model is developed using a Kaggle dataset consisting of 5,029 annotated CV samples. The research methodology includes dataset preparation, data preprocessing, tokenization, model fine-tuning, hyperparameter experimentation, and model evaluation. Multiple experiments are conducted by varying the hyperparameter configurations to observe their impact on the model's learning behavior and extraction performance. Experimental results shows that different hyperparameter settings could affect the performance of the model. The study identifies an optimal combination of hyperparameters that improves the model's ability to extract relevant information from CV documents dataset. Furthermore, the results demonstrate that hyperparameter tuning can provide better performance compared to the default BERT configuration, as reflected in improved evaluation metrics such as precision, recall, and F1-score. .
| Item Type: | Monograph (Project Report) |
|---|---|
| Subjects: | 000 Computer Science, Information and General Works > 004 Data processing & computer science |
| Divisions: | Faculty of Computer Science > Department of Informatics Engineering |
| Depositing User: | mr Dwi Purnomo |
| Date Deposited: | 11 Jun 2026 04:01 |
| Last Modified: | 11 Jun 2026 04:01 |
| URI: | http://repository.unika.ac.id/id/eprint/39993 |
| Keywords: | Natural Language Processing, Named Entity Recogniton, Information extraction, BERT-based Model, Hyperparameter Tuning, MUSLE |
Actions (login required)
 |
View Item |