01. 연구 (2) 원전 정책 담론 연구 02 연구 설계 및 구조

이 글은 이전 문서(00. 개요)의 후속 내용으로,

해당 연구 설계 및 구현 구조를 보다 구체적으로 정리했다.

 

블로그에 공개한 01–06 문서는 당시 컨택 과정에서 제출했던 18페이지 분량의 영어 연구계획서 가운데

방법론과 연구 구현에 해당한다. 

최종 제출본에는 분량 제한 때문에 일부만 반영되었지만,

방법론 부분이 가장 오래 수정 및 검토되었다.

 

연구계획서를 작성하는 일련의 과정 속에서,

나는 연구 과정을 스스로 설명할 수 있어야 한다는 점을 중요하게 여겼다.

 

예상과 다른 결과가 나오면 원인을 다시 확인했고, 구현 과정이 바뀌면 방법론도 함께 수정했다.

연구 질문과 분석 구조가 맞지 않는다고 판단한 부분은 여러 번 다시 설계했다.

 

그래서 블로그에 공개한 문서에는 완성된 결과보다 시행착오와 수정 과정, 

그리고 연구를 진행하면서 마주했던 한계와 고민이 더 많이 남아 있다.

 

본론

---

02. Project (Research) Overview

---

연구 규모가 커질수록 코드와 문서가 조금씩 따로 관리되기 시작했다. 데이터가 어떤 과정을 거쳐 분석되는지 스스로도 한눈에 파악하기 어려운 순간이 있었고, 전체 연구를 하나의 프로젝트 단위로 다시 정리하게 되었다. 이후에는 구현과 방법론을 수정할 때 기준이 되는 문서로 활용했다.

 

 

---

1994–2025 Integrated Analysis System for South Korean Radioactive Waste Policy and Discourse

01. Research Overview

---

• Period: 2026.03 – 2026.07
• Data: News Articles (BigKinds)
• Analysis Structure: Quantitative Analysis (Python + SQL-based data processing) + Qualitative Interpretation (Theoretical Framework)

---

Research Objectives

The objective of this study is to establish a research infrastructure that enables an integrated analysis of long-term changes in radioactive waste policy and related social discourse structures in South Korea based on news data.

Specifically, this study connects the timing of policy events with changes in news discourse to analyze the following:

1. Occurrence of discourse shifts following policy events
2. Emergence and evolution of discourse related to international norms
3. Structural changes in safety and social acceptance discourse
4. Time lags between policy events and discourse shifts

This study minimizes dependence on single analytical settings by structurally separating query design, data reduction, and keyword extraction stages.

---

02. Theoretical Background

International Norms

Organizations such as the IAEA and OECD-NEA provide policy standards. However, these are not directly reflected in news discourse but are mediated through policy events.

This study categorizes international norm-related keywords separately and analyzes their emergence patterns before and after policy events.

---

Policy Diffusion

International norms function as reference points for policy comparison, which domestic policies selectively adopt.

This study tracks changes in the frequency and importance of keywords related to international organizations to examine temporal relationships.

---

Discourse Analysis

• News Articles = Unit of Observation
• Keywords = Discourse Variables
• Frequency Change = Discourse Intensity

Discourse change is measured not only by frequency but also by TF-IDF-based relative importance, reflecting shifts in the concentration of discourse.

News data are used as observational data of public discourse. However, they do not directly represent overall public perception; rather, this study focuses on discourse structures mediated through media.

---

Integrated Analytical Perspective

This study interprets the policy–discourse relationship as co-variation, not causation.

The focus is on patterns of co-variation rather than causal inference.

---

03. Research Design

Analytical Framework

Multi-stage Data Reduction & Concept Validation Framework

This framework separates the roles of each stage to prevent any single step from determining the overall results.

---

Research Infrastructure

Pipeline structure:

Data Collection → Preprocessing → Storage → Analysis

• SQL: Structuring and relationship definition
• Python: Analysis and preprocessing
• Web: Visualization support

System implementation is treated as part of the analytical infrastructure.

This infrastructure was designed with scalability in mind from the initial stage.

It supports:

• Temporal expansion: Extension of the analysis period for long-term time-series studies
• Comparative analysis: Extension of policy event structures for cross-national or cross-policy comparison
• International norm linkage: Integration of analytical structures related to global standards
• Discourse category expansion: Extension of keyword categories to accommodate diverse discourse types

Accordingly, this structure is not limited to a single-case analysis, but provides a scalable analytical framework for examining policy–discourse relationships over time.

---

Keyword Structure

Type Role

Query Keywords	Data collection criteria
BigKinds Keywords	Metadata
TF / Weight / Level	Data reduction
Core Keyword	Conceptual analysis unit

---

TF Definition

TF is divided into two stages:

1. Data Reduction TF: Based on BigKinds metadata (for filtering)
2. Analysis TF: Corpus-based TF-IDF calculation

These are different in purpose and calculation and are not treated as identical.

---

04. Data and Methodology

---

4.1 Data Collection

• BigKinds
• CSV extraction
• Initial organization in Excel

---

4.2 Overall Procedure

① Literature Review

→ Establish analytical framework and conceptual definitions

↓

② Query Design

→ Define data collection criteria and scope

↓

③ Data Collection

→ Acquire news articles via BigKinds

↓

④ Data Reduction

→ Select data based on TF, weight, and level

↓

⑤ Data Cleaning

→ Remove noise and transform the data into a structured format

↓

⑥ Core Keyword Extraction

→ Define discourse variables via intersection

↓

⑦ Corpus Construction

→ Construct the analytical dataset

↓

⑧ Analysis

→ Measure discourse intensity and relative importance

---

4.3 Query Design

This stage defines collection criteria for extracting relevant data.

The query is designed based on theoretical judgment and consists of:

• 1st-stage Recall
• 2nd-stage Precision

The second-stage query functions as a sampling frame, not as a variable generator.

Multiple keyword combinations are applied per policy event to reduce bias.

---

4.4 Data Reduction

This stage optimizes information density from the raw corpus.

• Method: Rank-based Filtering
• Criteria: TF + BigKinds weight + keyword level

Data Reduction is a quantitative filtering stage, while Core Keyword extraction ensures conceptual consistency.

Top-N thresholds are applied adaptively based on data size.

---

4.5 Data Cleaning

Data cleaning removes residual noise and transforms the dataset into an analyzable structure.

SQL

• Duplicate removal
• Missing value handling
• Relationship definition

Python

• Text preprocessing
• Variable generation
• Corpus transformation

---

4.6 Core Keyword Framework

1) Definition

Core Keywords are defined as the intersection between:

• Policy concept set (query keywords)
• Data expression set (reduced corpus)

---

2) Derivation and Validation

Derived by cross-referencing Python-generated candidates with researcher-defined keywords.

Final selection is confirmed through manual validation based on:

• Frequency threshold
• Semantic consistency
• Normalization rules

---

3) Concept Normalization

Performed through iterative review using:

• Literature
• Policy documents
• Corpus context

---

05. Analysis Modeling and Mechanism

1) Modeling Definitions

• News Articles: Observation units
• Core Keywords: Analytical variables
• Policy Events: Reference points
• Results: Identification of structural shifts through changes in keyword distribution and weights

---

2) Analysis Mechanism

Policy Event

→ Media Response

→ Keyword Visibility Change

→ Discourse Restructuring

Time lag is explored using short-term and mid-term windows.

---

06. Analysis Stages and Principles

1) Techniques

• TF-IDF Analysis
• Time-Series Analysis
• Event-based Comparison

---

2) Principles

• Procedural Integrity: Use only finalized corpus
• Interpretive Validity: Co-variation, not causality

---

07. Conclusion

This study presents a reproducible and structurally stable policy discourse analysis framework based on:

• TF–Weight–Level integration
• Alignment of policy concepts and data expressions

---

08. Defense Logic

Criticism Response

Keyword arbitrariness	Normalization + intersection + criteria
TF inconsistency	Stage separation
Reproducibility	Python automation
Lack of causality	Co-variation definition
Manual intervention	Controlled validation
Query subjectivity	Recall/Precision separation
Reduction arbitrariness	Adaptive thresholds

---

09. Synthesis

This study proposes a structural analytical framework that derives Core Keywords through:

• Data reduction (TF–Weight–Level)
• Concept normalization
• Intersection-based alignment

By separating functional stages, the study ensures:

• Methodological consistency
• Structural stability

---

For a detailed explanation of the analytical procedures and methodological framework, refer to the

Research Methodology

document.

For the system architecture and data infrastructure design, see the

Research Infrastructure

document.