Abstract

This research paper presents a theoretical approach to sentiment-aware deep learning for maritime legal decision-making by using a multi-modal model. The approach is presented in the context of maritime law decision-making. The information contained in maritime legal documents, both in textual and graphical form, are brought together in the model that has been proposed to improve the accuracy and effectiveness of the decision-making process. We investigate cutting-edge algorithms such as BERT, GPT-3 and Roberta for textual sentiment analysis and pre-trained convolutional neural networks (CNNs) for extracting visual features. It is suggested that a fusion mechanism be used, such as attention-based or gated fusion, to integrate the multi-modal features efficiently. The paper also discusses interpretability and explains the ability of the proposed model, comparing it to baseline methods and analyzing its potential application in real-world maritime legal decision-making scenarios. In addition to this, the paper examines the baseline methods and compares the proposed model to them.

Keywords:BERT; GPT-3

Abbreviations:CNNs: Convolutional Neural Networks (CNNs); RNNs: Recurrent Neural Networks; LSTM: Long Short-Term Memory; GRUs: Gated Recurrent Units