Pilny, Andrew

Owned by UKY CCS RCD
Jun 09, 2024
2 min read

Project list

Transformer-Based Models for Coding Interaction in Work Group Meetings

Faculty PI: Andrew Pilny (Associate Professor, Department of Communication)

Objective: To enhance the automatic coding of dialogues in professional meetings by applying advanced transformer-based models to accurately classify conversational exchanges.

 

Computational Methods

The computational methods we will employ involve the use of Python programming language. Specifically, we will utilize the Hugging Face library, a state-of-the-art platform that provides pre-trained transformer models such as BERT and Llama 2.

These transformer models, thanks to their self-attention mechanisms, are particularly adept at understanding the context and nuances of human language and will be used to classify dialogue acts in our corpus of work group meetings. Both Python and the Hugging Face library are widely available and can be accessed freely. The transformer models from Hugging Face are open-source and commercially used, with extensive documentation and community support, ensuring their accessibility for our research purposes at UKY.

Our approach will also involve fine-tuning these models on the MRDA corpus to optimize their performance for our specific dataset. This aspect of method development is currently in progress, with the intention of customizing the models to better understand and code the intricacies of conversation in a professional setting.

Software Utilization in Project

  1. Python: A high-level programming language renowned for its clear syntax and readability, which is widely used in scientific computing and machine learning.

  2. Hugging Face Transformers Library: An open-source library providing a vast collection of pre-trained transformer models for natural language processing tasks, including text classification, information extraction, question answering, and more.

 

The project will employ the following software Python packages and libraries:

  1. PyTorch: Utilized for its DataLoader, RandomSampler, and SequentialSampler classes for efficient data handling and batching during model training.

  2. Scikit-learn: Employed for its performance metrics such as accuracy_score and classification_report to evaluate model performance.

 

In addition, the project will use the following general-purpose libraries for programming support:

  1. random: A Python module that implements pseudo-random number generators for various distributions.

  2. NumPy: A library for the Python programming language, adding support for large, multi-dimensional arrays and matrices, along with a large collection of high-level mathematical functions to operate on these arrays.

  3. time: This module provides various time-related functions for Python.

  4. datetime: Supplies classes for manipulating dates and times in both simple and complex ways.

Center for Computational Sciences