Sentiment analysis of public response to measurable fishing capture policy using LDA and LSTM methods

Abstract

Illegal, unreported, and unregulated (IUU) fishing poses a significant threat by depleting fish stocks, damaging marine ecosystems, jeopardizing economic livelihoods, and undermining long-term environmental sustainability. To address this, the government has implemented a public policy of measured fishing within the blue economy framework. Given the involvement of numerous stakeholders, it is crucial for the government to gauge public sentiment through tweets on social media platforms to evaluate and refine the policy’s implementation for greater effectiveness. While the long short-term memory (LSTM) method for sentiment analysis is adept at handling text sequences and context, it struggles with capturing contextual semantic correlations. Conversely, the latent Dirichlet allocation (LDA) method excels in identifying these correlations and uncovering dominant topics. This study shows that integrating LDA for topic modeling with LSTM for sentiment analysis enhances overall performance, providing more accurate and comprehensive insights into public responses and identifying key topics discussed in social media tweets.