Development of GPT-LSTM based Sentiment Interpretable Neural Network and Application to the Financial Sentiment Adaptation

Abstract

When deploying deep neural networks (DNNs) to services related to the financial area, "computational cost" and "black-box nature of updated parameters in DNNs" can be critical issues. To solve this problem, we first propose a novel sentiment interpretable neural network called GPT-LSTM based Sentiment Interpretable Neural Network (GL-SINN). In addition, as an apllication of this study, we propose a domain word polarity conversion method called "Word-level Polarity Adaptation framework based on SINN (WPAS)", which is the method of sentiment domain adaptation in a cost effectibve manner.