Title Page
Contents
ABSTRACT 12
Chapter 1. Introduction 16
1.1. Background and motivation 16
1.2. Research objectives and scope 22
1.3. Thesis organization 28
Chapter 2. Background 29
2.1. Theoretical background 29
2.1.1. Context information 29
2.1.2. Business analytics 34
2.1.3. Technology and product planning 36
2.2. Methodological background 45
2.2.1. Word embedding 45
2.2.2. Linguistic pattern mining 47
2.2.3. Sentiment analysis 50
2.2.4. F-term system 52
2.2.5. Link prediction 55
2.2.6. Deep neural networks 57
Chapter 3. Context-aware customer needs identification for new product planning 60
3.1. Chapter introduction 60
3.2. Proposed methodology 63
3.2.1. Preparing data using web scrapping and sentiment analysis 64
3.2.2. Extracting expressions using linguistic pattern mining 66
3.2.3. Defining clusters using BERT and k-means clustering 70
3.2.4. Identifying customer needs using co-occurrence analysis 72
3.3. Case study: Amazon Echo series 73
3.3.1. Collecting and preprocessing data 74
3.3.2. Extracting context expressions and product function expressions 74
3.3.3. Clustering context expressions and product function expressions 77
3.3.4. Identifying customer needs 82
3.4. Discussion 84
3.4.1. Novelty of the proposed approach 84
3.4.2. Applicability of the proposed approach 85
3.5. Chapter summary 88
Chapter 4. Discovering technology opportunities using technology-context information 91
4.1. Chapter introduction 91
4.2. Proposed methodology 94
4.2.1. Constructing a universal F-term network using co-occurrence analysis 96
4.2.2. Inserting artificial node to define firm's technology portfolio 97
4.2.3. Discovering technology opportunity using link prediction 99
4.2.4. Analyzing R&D directions using opportunity portfolio map 101
4.3. Case study: SHIMANO 105
4.3.1. Validation of proposed approach to the target firm 106
4.3.2. Suggestions for the target firm's R&D directions 116
4.4. Discussion 120
4.5. Chapter summary 123
Chapter 5. Predicting technology applicability considering organization-context information 127
5.1. Chapter introduction 127
5.2. Data preparation 130
5.2.1. Technology assignment database 130
5.2.2. Global research identifier database 132
5.2.3. Patent indicators 133
5.3. Experimental results 137
5.3.1. Identifying utilized and non-utilized UPRI patents 138
5.3.2. Extracting patent indicators 139
5.3.3. Building a UPRI technology-applicability prediction model 141
5.3.4. Evaluating the model performance 145
5.4. Discussion 147
5.4.1. The differences between UPRI and private firm technologies 147
5.4.2. Assessing the applicability of UPRI technologies 151
5.4.3. Organization-context information affects the applicability of UPRI technologies 154
5.5. Chapter summary 158
Chapter 6. Concluding remarks 161
6.1. Summary 161
6.2. Contributions 164
6.3. Future research directions 166
References 168
Abstract (in Korean) 192
Table 2-1. Summary of TOD studies based on target firm's current technology interests 41
Table 3-1. Different Stanford typed dependencies used for extracting context and product function expressions 69
Table 3-2. Algorithm for extracting context and product function expressions 75
Table 3-3. Examples of extracted context and product function expressions 76
Table 3-4. List of 37 context information clusters 79
Table 3-5. List of 47 product function clusters 80
Table 3-6. Top 20 identified customer needs 83
Table 4-1. Description of the terms used in this study 95
Table 4-2. Examples of F-term theme codes and viewpoints 110
Table 4-3. Opportunity F-terms within four technology-context information 111
Table 4-4. Some of the top 100 opportunity F-terms with their high link prediction scores 112
Table 4-5. Examples of technical deployments 115
Table 4-6. A patent search query for validation 122
Table 4-7. Validation results using Shimano patents applied from 2019 to 2023 123
Table 5-1. 26 patent indicators for predicting applicability of UPRI's patents 134
Table 5-2. Examples of extracted patent indicators 140
Table 5-3. Part of prediction results for the test dataset 143
Table 5-4. Confusion matrix of the trained model 145
Table 5-5. Performance evaluation of the trained model 145
Table 5-6. Differences between UPRI and private firm technologies 149
Table 5-7. Top 10 important patent indicators and technological dimensions 155
Figure 1-1. Business analytics framework 18
Figure 1-2. Context-based business analytics for product and technology planning 21
Figure 1-3. Scope of this dissertation 24
Figure 2-1. Embodiment of context information 32
Figure 2-2. Types of business analytics 35
Figure 2-3. Graphical example of Stanford typed dependencies for the sentence 49
Figure 2-4. Multi-dimensional subject indexing of F-term 54
Figure 2-5. F-term structure with views and their technical deployment 55
Figure 2-6. Concept of deep neural networks 58
Figure 3-1. Overview of the analysis framework 64
Figure 3-2. Exemplary review paragraph structure 65
Figure 3-3. Exemplary word embedding and clustering 70
Figure 3-4. Defining customer needs using co-occurrence of context information cluster and product function cluster 72
Figure 3-5. Gap statistic values: (a) context expressions; (b) product function expressions 78
Figure 3-6. Example of customer needs oriented from the context information and product function clusters 86
Figure 4-1. Overview of the proposed approach 95
Figure 4-2. Concept of universal F-term network construction 97
Figure 4-3. Construction of firm-centered F-term network 98
Figure 4-4. Opportunity F-term portfolio map 104
Figure 4-5. Overall procedure of the case study 106
Figure 4-6. Firm-centered F-term network: red=AFN, orange=held F-terms, and gray=candidate F-term 107
Figure 4-7. Performance comparison of link prediction algorithms (A: Precision @K, where k denotes the first k opportunity F-terms; and B: ROC-AUC) 108
Figure 4-8. Opportunity F-term portfolio map for the period 2010-2014 113
Figure 4-9. Opportunity F-term portfolio map for the period 2014-2018 117
Figure 5-1. Diagram of 12 recorded assignments for patent US 5216281 (Own representation based on Marco, Myers, Graham, D'Agostino, and Apple (2015)) 131
Figure 5-2. Process of developing a UPRI technology-applicability prediction model 138
Figure 5-3. Stratified K-fold cross-validation method (K=10) with hybrid sampling 142
Figure 5-4. Performance comparison of the proposed model and baseline models 146
Figure 5-5. Precision @K score of the proposed model and baseline models (K=200) 153
Figure 5-6. Partial dependence plot of top 4 patent indicators 158
Equation (4-1) A modified link prediction score 100
Equation (4-2) A heterogeneity score 103
Equation (5-1) A Binary cross-entropy loss 142
Equation (5-2) A precision at K score 152
Equation (5-3) A partial dependence function 156
Equation (5-4) A Monte Carlo method 156