This document serves as the technical README for BERTNN, a tool for predicting affective meanings using contextual embeddings. It's intended for users with a background in machine learning and natural language processing, who are interested in the technicalities of the BERTNN model.
BERTNN is built on the premise of leveraging the BERT (Bidirectional Encoder Representations from Transformers) model for sentiment analysis and emotional linguistics within a sociological context. The model's core functionality is to estimate Evaluation, Potency, and Activity (EPA) values of words in varying contexts, guided by MABMO grammar.
- BERT Architecture: BERTNN utilizes the transformer architecture from Hugging Face's Transformers library. The model is fine-tuned for the specific task of extracting affective meanings from text.
BERTNN's implementation involves several key components:
- Contextual Analysis: The model considers the context in which words are used to accurately estimate their affective meanings.
- EPA Value Estimation: Using BERT's deep learning capabilities, BERTNN estimates the EPA values for words in different roles (identity, behavior, modifier) within an event.
For researchers and practitioners, BERTNN offers several advanced features:
- Custom Event Analysis: Ability to analyze custom events crafted in MABMO grammar.
Contributions to BERTNN are welcome. Please adhere to the project's contribution guidelines for submitting pull requests or issues.
If you use BERTNN in your research, please cite:
Mostafavi, M., Porter, M. D., & Robinson, D. T. (2024). Contextual Embeddings in Sociological Research: Expanding the Analysis of Sentiment and Social Dynamics. [To be published ib] Sociological Methodology Journal.
Disclaimer: The BERTNN project is subject to ongoing development and improvements. The methods and data described in this documentation are liable to change.