작업은 얼마 안걸립니다. 적당히 읽고 추가명령 or Regenerate, 이후 Ctrl C+V 해서 Diagram 생성 전까지 5분 정도 걸렸습니다. ( 5분도 안걸립니다 )
1. 해결 주체 선정
해결 주체부터 선정해보겠습니다. 스스로 설정하느라 고민하지 마시고, 질문하세요.
전산 언어학자가 괜찮아보여서, 바로 Stop하고 진행했습니다.
2. 작업을 위한 워크플로우 생성
3. 워크플로우 정제
위 워크플로우는 ChatGPT가 작업하기에 크게 좋지 않습니다. ChatGPT가 작업하기 좋도록 Step by Step으로 작동할 수 있는 워크플로우를 만들어 달라고 요청하겠습니다.
하지만 이것도 너무 깁니다. 요구사항을 추가해보겠습니다.
4. 워크플로우 정제 2
나름 짧고 체계적인 단계를 만들었습니다. 이 정도면 어느정도 준비가 끝난것같습니다.
5. 프롬프트 생성
[ 역할 ] → [ 업무 ] → [ 업무 속 지침 ] 으로 생성합니다.
혹시 사용하실 분들을 위해 이부분은 복사할 수 있도록 작성하겠습니다.
From now on, you are a "computational linguist". Your task is to create a diagrammatic representation of the **article** I have presented to you. Please consider these instructions.
[Step 1 - Prepare the text: Open the article in a text editor.
Step 2 - Segment the text: Using a natural language processing tool like NLTK, break the text into sentences.
Step 3 - Parse the text: Use the same tool to perform POS tagging and dependency parsing to understand the structure of the sentences.
Step 4 - Identify important entities and topics: Apply NER to identify important entities. Use algorithms like LDA to discover key topics.
Step 5 - Summarize the text: Use a tool like BERTSum to create a summary of the text.
Step 6 - Create a hierarchy: Based on your understanding of the text, create a hierarchy with the main topic at the top and supporting subtopics and details.
Step 7 - Design a mermaid diagram: Start by designing a diagram in the mermaid scripting language. For example, the main topic could be the starting point for the diagram. The first level nodes become subtopics, and then other details are added.
Step 8 - Refine the mermaid diagram: Adjust and refine the mermaid diagram as needed to best represent the logical structure of your text.
Step 9 - Review: Review the entire process and diagram. Make any necessary changes and improvements based on your review.
Step 10 - Finalize and render: When you are satisfied, finalize the mermaid diagram and render the diagram using the mermaid live editor or similar tool.
Provide only the final product].
Answer "OK" to confirm and I will paste the **Article**.
6. 2차 요청
이후 2차 요청을 하였습니다.
Using the provided article and following the described steps for analysis and diagram creation, here's a potential representation in Mermaid.js:
mermaid
Copy code
graph TB A["The Japanese culture of having a place for everyone"] --> B["Origin: Japan's traditional social and cultural system"] A --> C["Comparison: Korea's Confucian and patriarchal culture"] A --> D["Concept: Acting within one's own class and family traditions"] D --> E["In-depth: Acting in one's own fraction - Ruth Benedict"] D --> F["Impact: Invisible sense of class in modern democratic society"]
In this diagram:
A is the main topic.
B, C, and D are subtopics derived from the article.
E and F are details about the subtopic D.
7. 결과 비교
Prompt 미사용 ( Show me Plugin 사용 )
Prompt 사용 후
차별점
1. 문장의 설명 방식을 키워드 하나로 정의 ( ex. 비교, 개념... ) 2. 문장간 연관성에 따라 Grouping 시도 3. Regenerate 시에도 이전보다 안정적인 결과값 도출
저는 이정도면 발표 대본으로 쓰기에는 충분한 Diagram 인 것 같습니다. 위 Diagram을 바탕으로 쓰면서 중요한것만 더 체크하면 될것같아요.
8. Regenerate에 대한 시도
Prompt의 정제도(얼마나 과정이 명확하고, Output Format을 잘 지시했는지)에 따라 Regenerate시의 결과값이 달라집니다. 결과값이 명확하게 지정된 Prompt일수록 더 일정한 Output을 줍니다. 하지만 일정한 Output을 주는 Prompt가 항상 좋은건 아닙니다. 내가 원하는 결과를 가져다 주는게 제일 좋은 Prompt입니다.
내가 원하는 Output이 나올 때 까지 반복해서 실행해보세요. ( 물론 몇번해서 안나오면 Prompt를 수정해야 합니다. )
그리고 내가 원하는 Output이 나왔다면, 그 Output값을 다시 입력해주세요. 그럼 이전보다 더 일관되게 작업을 수행해줍니다.