Important Note: This is the old version of the code for this project. The latest version of the code will be released together with the relevant research paper at the following URL after all debugging is complete and the paper is submitted: https://github.com/mingfenghong/llm_assisted_grounded_theory

This project aims to use Large Language Models (LLMs) to assist with Grounded Theory analysis. The entire workflow uses "Open Source Cases" as the subject of study. Through several automated stages, it extracts causal relationships from the original case texts and progressively builds a theoretical model. The project's code is clearly divided into five core stages, from data extraction to the theory saturation test.

This project connects five core stages through main.py, with each stage calling the DeepSeek API to perform specific analysis tasks:

1. Stage 1: Information Extraction

   • Script: stages/stage1_information_extraction.py
   • Task: Reads the text of AI model open-source cases from the input Excel file (test.xlsx) and extracts causal relationship pairs.
   • Output: Generates step1.xlsx, which includes the case number, causal pair number, original text of the cause, original text of the effect, and the source text snippet.

2. Stage 2: Open Coding

   • Script: stages/stage2_open_coding.py
   • Task: Reads the causal pairs from step1.xlsx and performs open coding on the "cause" and "effect" to generate concepts and subcategories.
   • Output: Generates step2.xlsx, adding cause_concept, effect_concept, cause_subcategory, and effect_subcategory fields to the existing data.

3. Stage 3: Axial Coding

   • Script: stages/stage3_axial_coding.py
   • Task: Extracts all subcategories from step2.xlsx and consolidates them into main categories through cluster analysis, providing explanations for each.
   • Output: Generates step3.xlsx, which contains the main categories, their explanations, and a list of subcategories under each main category.

4. Stage 4: Selective Coding

   • Script: stages/stage4_selective_coding.py
   • Task: Identifies the core category from the main categories in step3.xlsx and elaborates on its relationship with other main categories.
   • Output: Generates step4.xlsx, containing the core category and its explanation.

5. Stage 5: Data Saturation Test

   • Script: stages/stage5_saturation_test.py
   • Task: Uses data from step2.xlsx and step3.xlsx to test whether all subcategories can be covered by the existing main categories. If not, it proposes new main categories.
   • Output: Generates step5.xlsx, which shows the classification label for each subcategory or proposes new categories.

The prompts used for analysis in each stage of this project are from the following research:

Zhou, Y., Yuan, Y., Huang, K., & Hu, X. (2024). Can ChatGPT perform a grounded theory approach to do risk analysis? An empirical study. Journal of Management Information Systems, 41(4), 982-1015.

/
|-- main.py                     # Main program entry point
|-- data/
|   |-- input_data/
|   |   `-- test.xlsx           # Raw input data
|   |-- output_data/            # Output results for all stages
|   |   |-- step1.xlsx
|   |   |-- step2.xlsx
|   |   |-- step3.xlsx
|   |   |-- step4.xlsx
|   |   `-- step5.xlsx
|   `-- data_processing.py      # Utility functions for data reading, writing, and processing
|-- stages/                     # Scripts for each analysis stage
|   |-- stage1_information_extraction.py
|   |-- stage2_open_coding.py
|   |-- stage3_axial_coding.py
|   |-- stage4_selective_coding.py
|   `-- stage5_saturation_test.py
|-- api/
|   `-- deepseek_client.py      # Client that encapsulates DeepSeek API calls
`-- README.md                   # This file

1. Configure Environment:

   • Install the required Python libraries, mainly pandas and openai.
   • Configure your DeepSeek API key in the api/deepseek_client.py file.

2. Prepare Data:

   • Organize your raw case data into the data/input_data/test.xlsx file. This file should contain at least two columns: Number of Cases and Case text.

3. Run Analysis:

   • Run the main program main.py.

     python main.py

   • The program will execute all five stages in sequence and generate intermediate and final result files in the data/output_data/ directory.

• pandas
• openai

重要提示： 这是本项目的旧版代码。最新版本的代码将在全部调试完成与论文提交后随相关研究论文一同发布在以下网址：https://github.com/mingfenghong/llm_assisted_grounded_theory

本项目旨在使用大型语言模型（LLM）辅助进行扎根理论（Grounded Theory, GT）分析。整个流程以“开源案例”作为研究对象，通过多个自动化阶段，从原始案例文本中提取因果关系，并逐步构建理论模型。项目代码清晰地划分了从数据提取到理论饱和度检验的五个核心阶段。

本项目通过 main.py 串联起五个核心阶段，每个阶段都调用 DeepSeek API 来完成特定的分析任务：

1. 阶段1：信息抽取 (Stage 1: Information Extraction)

   • 脚本: stages/stage1_information_extraction.py
   • 任务: 从输入的 Excel 文件（test.xlsx）中读取AI模型开源案例的文本，提取因果关系对。
   • 输出: 生成 step1.xlsx，包含案例编号、因果对编号、原因原文、结果原文以及来源文本片段。

2. 阶段2：开放编码 (Stage 2: Open Coding)

   • 脚本: stages/stage2_open_coding.py
   • 任务: 读取 step1.xlsx 中的因果对，对“原因”和“结果”进行开放编码，生成概念（Concept）和子范畴（Subcategory）。
   • 输出: 生成 step2.xlsx，在原有数据基础上增加了 cause_concept, effect_concept, cause_subcategory, effect_subcategory 字段。

3. 阶段3：主轴编码 (Stage 3: Axial Coding)

   • 脚本: stages/stage3_axial_coding.py
   • 任务: 从 step2.xlsx 中提取所有子范畴，通过聚类分析将其归纳为主范畴（Main Category），并提供解释。
   • 输出: 生成 step3.xlsx，包含主范畴、其解释以及该主范畴下的子范畴列表。

4. 阶段4：选择性编码 (Stage 4: Selective Coding)

   • 脚本: stages/stage4_selective_coding.py
   • 任务: 基于 step3.xlsx 中的主范畴，识别出核心范畴（Core Category），并阐述其与其他主范畴之间的关系。
   • 输出: 生成 step4.xlsx，包含核心范畴及其解释。

5. 阶段5：理论饱和度检验 (Stage 5: Data Saturation Test)

   • 脚本: stages/stage5_saturation_test.py
   • 任务: 使用 step2.xlsx 和 step3.xlsx 的数据，检验所有子范畴是否能被已有的主范畴所覆盖。如果不能，则提出新的主范畴建议。
   • 输出: 生成 step5.xlsx，展示每个子范畴的归属标签或新范畴提议。

本项目中用于各阶段分析的提示词（Prompts）主要来自以下研究：

Zhou, Y., Yuan, Y., Huang, K., & Hu, X. (2024). Can ChatGPT perform a grounded theory approach to do risk analysis? An empirical study. Journal of Management Information Systems, 41(4), 982-1015.

/
|-- main.py                     # Main program entry point
|-- data/
|   |-- input_data/
|   |   `-- test.xlsx           # Raw input data
|   |-- output_data/            # Output results for all stages
|   |   |-- step1.xlsx
|   |   |-- step2.xlsx
|   |   |-- step3.xlsx
|   |   |-- step4.xlsx
|   |   `-- step5.xlsx
|   `-- data_processing.py      # Utility functions for data reading, writing, and processing
|-- stages/                     # Scripts for each analysis stage
|   |-- stage1_information_extraction.py
|   |-- stage2_open_coding.py
|   |-- stage3_axial_coding.py
|   |-- stage4_selective_coding.py
|   `-- stage5_saturation_test.py
|-- api/
|   `-- deepseek_client.py      # Client that encapsulates DeepSeek API calls
`-- README.md                   # This file

1. 配置环境:

   • 安装所需的 Python 库，主要是 pandas 和 openai。
   • 在 api/deepseek_client.py 文件中配置您的 DeepSeek API密钥。

2. 准备数据:

   • 将您的原始案例数据整理成 data/input_data/test.xlsx 文件。该文件应包含至少两列：Number of Cases（案例编号）和 Case text（案例文本）。

3. 执行分析:

   • 运行主程序 main.py。

     python main.py

   • 程序将按顺序执行所有五个阶段，并依次生成中间和最终结果文件到 data/output_data/ 目录下。

• pandas
• openai