Khóa luận tốt nghiệp: Nhận dạng thực thể trong văn bản tiếng Việt sử dụng mô hình học sâu suốt đời mức ký tự

In the realm of Artificial Intelligence and data analysis, Machine Learning (ML) and particularly Deep Learning (DL) have emerged as pivotal tools. However, conventional supervised ML methods face significant challenges, such as the high cost and time commitment of acquiring large labeled datasets, and the need to continuously adapt to evolving information. This necessitates approaches that can learn incrementally and retain knowledge over time. Lifelong Machine Learning (LML), which mimics human cognitive processes by accumulating and transferring knowledge from past tasks to new ones, offers a compelling solution. This thesis explores the integration of deep learning and lifelong learning to address the critical task of Named Entity Recognition (NER), particularly within the context of Vietnamese text, paving the way for more robust and adaptive NLP systems.

Researchers, academics, and students in Natural Language Processing (NLP), Machine Learning (ML), Deep Learning (DL), and Artificial Intelligence (AI), particularly those interested in information extraction for low-resource languages like Vietnamese.

This thesis addresses the complex problem of Named Entity Recognition (NER) in Vietnamese text by proposing an innovative approach that combines deep learning (DL) and lifelong machine learning (LML). The research delves into understanding and integrating these two powerful paradigms. Specifically, it focuses on developing a character-level deep lifelong learning model tailored for NER in Vietnamese. The methodology involves examining existing deep learning models like Bi-LSTM-CRF, which are effective for sequence labeling tasks, and adapting lifelong learning principles for continuous knowledge acquisition and transfer. The proposed model is designed to overcome the limitations of traditional models, which often struggle with data scarcity and the inability to leverage previously learned knowledge for new tasks. To validate the model's efficacy, comprehensive experiments were conducted using the VLSP2018 dataset, a standard benchmark for Vietnamese NLP, and an additional dataset collected from the Dân trí online newspaper for lifelong feature extraction. The experimental results demonstrate the promising performance and effectiveness of the character-level deep lifelong learning model. This work not only contributes to advancing NER capabilities for Vietnamese, a language often considered low-resource in NLP research, but also highlights the significant potential of combining deep learning with lifelong learning to create more intelligent, adaptive, and efficient natural language processing systems that can continuously learn and improve over time. The implications extend to various information extraction tasks, setting a foundation for future research in robust AI applications.