Skip to content

train a generative model based on persuasion strategies. use another model to classify strategies in the training dataset, and compare the accuracy on classifying the generated data.

Notifications You must be signed in to change notification settings

moonsujo/persuasion-conversation-generator

Repository files navigation

ACGenV: Automated Conversation Generation & Evaluation

This repository stores all code used for our project in CS544 Fall 2022.

There are two main elements:

  • conversation-generator*
  • conversation-analyzer

*the model(s) used in the conversation-generator were not able to be uploaded entirely to GitHub due to their large size and can be made available through a Google Drive link upon request.

There are also two evaluation models:

  • Perplexity
  • BLEU Score

Perplexity and BLEU scores are generated wrt subsets of the original PSG dataset, and are compared to both of our generative conversational models' output. Details of the BLEU Score implementation can be found in the bleu_scores folder and the Perplexity metric implementation in the perplexity_metric folder.

About

train a generative model based on persuasion strategies. use another model to classify strategies in the training dataset, and compare the accuracy on classifying the generated data.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published