Added level_0 CSV dataset

[rowbotham-evan]

Notes

• Generated a synthetic dataset that measures construct validity (referred to as "level_0" internally)
  • Construct validity tests the measurement (i.e. how well do our features measure/align with our predictions based on "perfect" and vacuous data – specifically at the sentence level)
  • Generates 50 conversations of each type:
    • high directness, high OI
    • high directness, low OI
    • low directness, high OI
    • low directness, low OI
• Generated a synthetic dataset that measure ecological validity (referred to as "level_1 internally"
  • Ecological validity tests allow us to establish that these features correlate to predictions make sense at the conversational level. (i.e. does the sentence we are measuring in a generated conversation go awry or is productive)
  • Generates 100 awry (escalatory conversations) and 100 productive (de-escalatory conversations)
• To generate another "level 1" or "level 0"dataset reference the GPT_API_script in Priya's conflict coefficientofconflict repo