Difference between revisions of "Research Goals"

From REU@MU
Jump to: navigation, search
 
(3 intermediate revisions by the same user not shown)
Line 1: Line 1:
 
'''RQ1''': How do medical crowdfunding campaigns differ from AI-generated replications?
 
'''RQ1''': How do medical crowdfunding campaigns differ from AI-generated replications?
 
* Step 1: Scrape GoFundMe's [https://www.gofundme.com/start/medical-fundraising medical] fundraising homepage. I should be able to adapt Alex Fischmann's script from last summer.
 
* Step 1: Scrape GoFundMe's [https://www.gofundme.com/start/medical-fundraising medical] fundraising homepage. I should be able to adapt Alex Fischmann's script from last summer.
* Step 2: Prompt an LLM (GPT-3 or GPT-4) to rewrite scraped campaigns. This may require fine-tuning. Write a script to replicate ''n'' campaigns ''n'' times. We want a 1-1 relationship between an original campaign and its replication. In other words, the output should adhere to the facts in the prompt.
+
* Step 2: Prompt an LLM (GPT-3 or GPT-4) to rewrite scraped campaigns. This may require fine-tuning. Write a script to replicate ''n'' campaigns ''n'' times. We want a 1-1 relationship between an original campaign and its replication. In other words, the output should adhere to the facts provided in the prompt.
 +
* Step 3: Use natural language processing (NLP) techniques to compare original campaigns to their AI-generated counterparts. Analyze individual word associations with NRC or LIWC and overall tone with sentiment [https://nlp.stanford.edu/sentiment/ analysis].
 +
'''RQ2''': Which campaign is more successful with respect to medical crowdfunding?
 +
* We need IRB approval to conduct experiments with human participants.
 +
* Which factors contribute to a successful campaign? How do we quantify success?

Latest revision as of 23:56, 2 June 2023

RQ1: How do medical crowdfunding campaigns differ from AI-generated replications?

  • Step 1: Scrape GoFundMe's medical fundraising homepage. I should be able to adapt Alex Fischmann's script from last summer.
  • Step 2: Prompt an LLM (GPT-3 or GPT-4) to rewrite scraped campaigns. This may require fine-tuning. Write a script to replicate n campaigns n times. We want a 1-1 relationship between an original campaign and its replication. In other words, the output should adhere to the facts provided in the prompt.
  • Step 3: Use natural language processing (NLP) techniques to compare original campaigns to their AI-generated counterparts. Analyze individual word associations with NRC or LIWC and overall tone with sentiment analysis.

RQ2: Which campaign is more successful with respect to medical crowdfunding?

  • We need IRB approval to conduct experiments with human participants.
  • Which factors contribute to a successful campaign? How do we quantify success?