search

DDMAL / linkedmusic-datalake

To create mapping strategies for various music databases into our data lake
https://virtuoso.staging.simssa.ca
0 stars 4 forks source link

Prompt Engineering Study and Trial & LLM to SPARQL #49

Open candlecao opened 5 months ago

candlecao commented 5 months ago

1) Start with https://platform.openai.com/docs/guides/prompt-engineering. 2) Figure out basic concepts and questions. 3) An appointment: Kun Fang gives us an illustration about prompt engineering and how to use the API of GPT(in the morning, May 29). 4) With basic prompt engineering skill&understanding, further review literatures about LLM2SPARQL. 5) Connection with LinkedMusic's specific prompt.

LLM is short for Large Language Model.

candlecao commented 5 months ago
  1. Connection with LinkedMusic's specific prompt: -> For SimssaDB: I used ChatGPT 3.5 to convert 4 natural language queries into (1)SPARQL (2)the outcome from the SPARQL in format like spreadsheet (3)the explanation on the outcome.
candlecao commented 5 months ago

Hi Ich, @fujinaga please take a look at the branch "PromptEngineering_on_simssaDB" that is about LLM2SPARQL: Using Gpt 3.5, do an experiment on SimssaDB sample: 1) Ask five question_1-5(from easy to hard). 2) Provide a context to help GPT understand. 3) Program with Python. 4) Document the result. Now the context is just sufficient and not superfluous so as to let GPT generate right SPARQL result with higher probability.

Your comments and guidance will be greatly appreciated. Thanks.