Title Page
Abstract
국문 초록
Preface
Contents
Chapter 1. Introduction 15
Chapter 2. Related Work 19
Chapter 3. Simple and Efficient Data Augmentation 22
3.1. Synonyms/Antonyms Based Data Augmentation 23
3.2. POS-tagging and Lemmatizing Based Data Augmentation 25
3.2.1. Random and Brute Data Augmentation 25
3.2.2. POS-tagging Based Data Augmentation 25
3.2.3. Lemmatizing Based Data Augmentation 28
Chapter 4. Model 32
Chapter 5. Experiments 36
5.1. Dataset Details 36
5.2. Implementation Details 37
5.3. Evaluation Details 38
5.4. Main Results 38
5.5. Performance Comparison Experiment According to the Amount of Data 40
Chapter 6. Conclusion 42
Bibliography 43
Table 1.1. Performance of TOEIC-BERT, an existing study that conducted experiments with a deep learning model for TOEIC Part 5. 16
Table 3.1. Examples of TOEIC Part 5 fill-in-the-blank in a single sentence. 22
Table 3.2. Parts-of-speech tags included in this dataset. 27
Table 3.3. Examples of POS-tagging Based Augmented Data and Lemmatizing Based Augmented Data. 29
Table 5.1. The statistics of the training, validation, and test set. Here, the 'length' refers to the length after tokenization with BertTokenizer. 36
Table 5.2. Experimental results on Kaggle TOEIC data, Synonyms/Antonyms Based Augmented Data, Randomly and Brutely Augmented Data, POS-tagging Based Aug-... 39
Figure 1.1. Distribution of the number of people by the score for the 501st test (implemented on 12 November 2023). 16
Figure 3.1. Overall process of Synonyms/Antonyms Based Data Augmentation Methodology 24
Figure 3.2. Overall process of POS-tagging and Lemmatizing Based Data Augmentation Methodology 31
Figure 4.1. Input and output process of the model. 35
Figure 5.1. BERT, RoBERTa, and ELECTRA performance variation according to the data size. 41