GPT Reviews

Installing Stopwords

In order to use stopwords, have to open Python interpreter

import nltk
nltk.download('stopwords')

Steps

Before I forget how this works, quick instructions: Main.py

• Need a distinct list of make / model / year / type from supabase

• Update to and from count in .env

• If you want to skip reviews, uncomment the thing (mainly for making critical reviews)

• review_generation.generate_review will run

• This is where ChatGPT lies

• Can set different word counts, use random_utils to set different topics

• Will need to change prompts for regular reviews vs critical

• Can set own topics to have it go off of (more topics the better)

• Note some topics have been kinda bad ("Just say it was okay","Just say you liked it","Just say it's a really great product") these will sometimes literally have those as the review

• NEED TO update folder names and/or file names for input, output, and reports

• After reviews have been generated, will attempt to parse the GPT reviews to put it into a structure for writing to CSV

• This parser will try to find Title and Content

  • However, there was a time where I was doing (Helpful: and Content:) and parsing it like that. GPT had issues being consistent about it, so might be better assigning it after

• After the parser, it should write a CSV

• If you have multiple CSV (because you batched it), run the csv_combine (you can run just the script, don't need to do it from main)

  • Have to update foldesr/filenames if needed

• After that, need to clean it, can run the script itself , don't need to do main

  • Have to update foldesr/filenames if needed

• After that, can upload it to supabase

  • Give it id column (uuid)

• Find the scripts for reviews

  • Probably have to add review at column to the gpt review table
  • randomize that
  • if images are in can skip that step in the sql sccript
  • insert into reviews table, make sure slugs are updated with the script

• If review images aren't already starting with it , have to create another csv with id, type, helpful and the count

• Run the fix_image_and_helpful script

• Add table to supabase

• Update the GPT table with that one

• and then update the rest of stuff

• And then i dno't know what the heck else there is to do

TODOS

• Improve the initial review generation to include cover type (will be able to remove a step)
• Probably also want to add in columns for review_image
• Include step for randomizing helpfulness and adding photos in an earlier step (currently it's away)
• Need to re-incorporate critical reviews with the process so don't have to do separate step
  • Will probably need some more ternaries and stuff -[x] Make changing folder names / file names easier
• Improve Supabase Script or extract the important parts out (GPT Reviews & Reviews-2)