Sentiment Analysis for Women in STEM using Twitter and Transfer Learning Models

Abstract

The science, technology, engineering and math (STEM) sector is integral to the nation’s advancement and economy. However, the STEM workforce is perceived as maledominant, and women are systematically tracked away from it. There has been a rising popularity of the gender disparity problem in STEM across social media platforms. Attitudes relating to women influence the careers women choose to pursue. It is thus timely and important to assess the public’s opinion on this topic. This paper proposes a sentiment analysis classification framework that detects the sentiment of social media tweets in relation to women in STEM. To this end, we extracted more than 250,000 relevant tweets and used various open-language models to uncover insights into the perceptions of women in STEM using various open-language models. The study evaluates the performance of multiple machine learning and deep learning methods. We also study the performance of state-of-the-art transformer based models, including bidirectional encoder representations from transformers (BERT), BERTweet, and TimeLMs (Time Language Models), which achieves 96% accuracy in sentiment detection. Results reveal that people’s attitude in response to women in STEM is generally positive on the Twitter platform. However, we observed a significant correlation between positive sentiment in tweets and dates celebrating women’s achievements (e.g. International Day of Women and Girls in Science, and International Women’s Day). This finding demonstrates the impact of such campaigns on the public’s opinion. Therefore, promoting these events among the public can encourage more females to pursue careers in STEM.