Title Page
ABSTRACT
국문 초록
Contents
Chapter 1. Introduction 15
1.1. Economics data analysis 15
Chapter 2. Keyword Extraction in Economics Literatures 19
2.1. Introduction 19
2.2. Related Works 20
2.2.1. Natural Language Processing 20
2.2.2. Bidirectional Encoder Representation from Transformer (BERT) 20
2.3. Data and Methods 24
2.4. Experiment Results 27
2.5. Discussion and Conclusion 28
Chapter 3. Relation Extraction and Knowledge Graph Construction of Economic Journals Paper 29
3.1. Introduction 29
3.2. Related Works 31
3.2.1. Relation Information 31
3.2.2. Knowledge Graph 32
3.2.3. BERT Language Model 34
3.3. Relation Extraction between Keyword Entities 35
3.3.1. Journal Paper Keyword Entities Extraction 36
3.3.2. Relation Information Extraction 38
3.3.3. Knowledge Graph Extraction 40
3.4. Experiment 42
3.4.1. Experiment Data and Materials 42
3.4.2. Experiment Environment 45
3.4.3. Experiments on Relation Extraction using BERT 46
3.4.4. Knowledge Graph Implementation of Relationship Information 49
3.5. Discussion and Conclusion 51
Chapter 4. Multi-label Text Classification of Economic Concepts from Economic News Articles 52
4.1. Introduction 52
4.2. Preliminaries 53
4.2.1. Multi-label Classification 53
4.2.2. Multi-label Text Classification 53
4.3. Proposed Method 55
4.4. Experiment Results 58
4.5. Discussion and Conclusion 63
Chapter 5. Conclusion 64
REFERENCES 65
Table 1. The Accuracy Result of Experiment 28
Table 2. Dataset Statistics 44
Table 3. Parameter Settings 45
Table 4. Example of the hypothesis in Figure 8 inferred from the economic paper data 46
Table 5. Predicted relationship classification values in Table 6 47
Table 6. Experiment Results 48
Table 7. Results of large-scale text data experiments[내용없음] 12
Table 8. The Accuracy Result of Experiment 62
Figure 1. Differences between Traditional ML and Transfer Learning 21
Figure 2. Transformer Encoder of BERT 22
Figure 3. Self-Attention 23
Figure 4. Work flow of keyword extraction from Economics journal paper using BERT 25
Figure 5. A train loss graph according to the epochs 27
Figure 6. Information Extraction Pipeline 37
Figure 7. Method: (a) BERT_CLS (b) BERT_Entity 37
Figure 8. Relation Graph Hypothesis 41
Figure 9. Graphs formed from Relation Information extracted from Economics Papers 50
Figure 10. Differences in Classification Tasks 54
Figure 11. The architecture of the proposed multi-label sentence-level event classification model 57
Figure 12. Pseudocode for the Multi-label Classifier Algorithm 58
Figure 13. Frequency of Word Counts per sentence 59
Figure 14. Train and Validation Loss Graph 60
Figure 15. Classification Report of True and Predicted labels 60
Figure 16. Example of sentence, its true labels and prediction labels 62