Title Page
Contents
Abstract 11
1. Introduction 13
2. Related Works 18
2.1. Graph Neural Networks 18
2.2. Graph Neural Networks for Recommender Systems 19
2.3. Cold Start Problem 21
3. Methods 22
3.1. Preliminaries 22
3.2. Creation of Heterogeneous Graph 23
3.3. Subgraph Extraction 24
3.4. Edge Labeling 24
3.5. Node Labeling 26
3.6. Learning Model 27
3.7. Feature Selection 30
4. Experiments 32
4.1. Dataset 32
4.2. Experiment Settings 32
4.3. Warm Start Test 33
4.4. Cold Start Test 36
5. Conclusions 39
References 40
ABSTRACT IN KOREAN 45
Table 4.1. Edge value determination experiment 33
Table 4.2. Performance Comparison of Hi-GMC by Feature 34
Table 4.3. Model-specific Weight Value for Ensembles 34
Table 4.4. Test results on MovieLens-100K 36
Figure 1.1. Examples of when content information is not used and used in movie... 15
Figure 3.1. Heterogeneous graph with node content example. All side information... 24
Figure 3.2. Example of rating copy. Simplified graph structure for understanding.... 25
Figure 3.3. Example of feature node labeling. The gender of the target user is female,... 27
Figure 4.1. RMSE for n% lower users 38
Figure 4.2. RMSE for sparsity ratio 38