A recent paper investigates the measurement of sentiment in relation to Environmental, Social, and Governance (ESG) across social media platforms. As the ESG framework gains traction in financial and business sectors, the demand for accurate sentiment analysis increases. The study employs human researchers to classify the sentiment of 150 tweets, which are then used to create a gold standard dataset. This dataset is benchmarked against various machine approaches, including the VADER dictionary approach and language models such as Llama2, T5, Mistral, Mixtral, FINBERT, and notably, GPT3.5 and GPT4. Read the full paper here.

• Measurement of ESG sentiment is growing in importance for business performance.
• Comparison study between human performance and language models.
• GPT variants, alongside other models, are analyzed for their sentiment classification accuracy.
• A gold standard sentiment dataset based on human consensus is created for benchmarking purposes.

This research is groundbreaking as it pits human insight against machine precision in the nuanced domain of ESG sentiment. In an era where corporate reputation is closely tied to ESG principles, these insights pave the way for more sophisticated sentiment analysis tools and could significantly impact financial and reputation metrics in businesses.