thomasschaller1/seminar-chatbot

Seminar Chatbot - Extracting the personality of a person from a chat conversation

The purpose of this project is to extract the personality of a person using Transformers on labeled text data of the Big Five model.

Getting Started

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes. See deployment for notes on how to deploy the project on a live system.

Prerequisites

What things you need to install the software and how to install them

conda create --name <env> --file requirements.txt
or
install pandas, numpy, pytorch, transformers, simpletransformers, (optional) ktrain

Usage

Data can be found in the data folder. There is 3 datasets:

Essays
FriendsPersonality (but finally not used)
myPersonality

Source code can be found in the src folder.

• Useful code, result used for the paper:
  • data.py: contains code to preprocess and format the data
  • simpletransformers_train.py: file used to train every model with different parameters. Almost all good results come from this file
  • simpletransformers.ipynb: Test of simpletransformers before implementation in the python file
  • Seminar_Chatbot_Demo.ipynb: File used for the demo, using the model of cCON and cEXT, on google colab.
  • main.py: self-made training using RoBERTa model
  • predict.py: predict data using model get from main.py
• Test code, finally not used for the paper:
  • ktrain.ipynb and ktrainv2.ipynb: Test of using the ktrain library
  • test1.ipynb, test2.ipynb, v1.ipynb, v2.ipynb: Test file of different model

To test the training, use simpletransformers_train.py and uncomment / change the different array at the beginning of the file. To test a model, use Seminar_Chatbot_Demo.ipynb and change the folder dir and the name of the model when creating ClassificationModel

Authors

• Thomas Schaller

License

This project is licensed under the MIT License

Acknowledgments

This is paper is a student work, written for Seminar "Chatbots and Conversational Agents" of the university of Fribourg, under the supervision of Jacky Casas and Prof. Dr. Elena Mugellini, HumanTech Institute, University of Applied Sciences of Western Switzerland, Switzerland, 2020.