Creating a Historical Sentiment Lexicon Using Johnson’s Dictionary and HPC
Project Overview
This project will built a historical sentiment lexicon using Johnson’s 1755 Dictionary with Google’s BERT model: the 50,000 headwords, definitions and examples will be used to calculate a sentiment score for each word, which can then be applied to historical texts. The aim is to try out different ways of developing scores and test the usefulness on articles from thousands of historical newspapers using HPC. Whilst there have been some small-scale efforts to build lexica specific to datasets, this project represents a major step forward in making sentiment analysis historically rigorous.
What were the key results of your research project?
This project successfully created a historically accurate sentiment lexicon that held up under rigorous training on the HPC, using different manually annotated batch sizes.
This allows sentiment to be gauged on historical texts against modern-day understanding and bias. Manual and human anecdotes were the most valuable annotations, but the BERT model did well to categorise, learn, and repeat language patterns with 92% accuracy.
This is the first time a historical lexicon has been created to allow us to delve into the reception of language outside of modern language exploration.
GitHub Repository: https://github.com/milliecain/sentiment-analysis
A YouTube video of this presentation will be available in the coming month.
How do you feel you have benefitted from completing this internship and has it made you consider future career paths?
I've really appreciated the opportunity to learn digital tools and computational methods from a research standpoint, having never had the support before. Both of my supervisors and the entire RSE team were fantastic and helped me progress my ideas and gave me the space to make mistakes and learn from them throughout the project.
Learning more about the role of an RSE has truly inspired me to look forward to digital humanities careers and has opened my eyes into another path of research, which I can take forward into my final year of study and the rest of my career.
Being able to learn Python, become familiar with the HPC and be surrounded by experts has been an incredible learning experience, and one I have learnt so much from, not just computationally, but thoughtfully.
Download slides of the presentation