Penalty based Sentimental Text Generation Framework using Generative Adversarial Networks

Emotional Intelligence is the technology where computers are trained to handle, use, understand and manage emotions. Emotional Intelligence (EI) is used in sentimental text generation where sentimental texts like positive, negative and neutral texts are generated. Sentimental text generation is used in many Artificial Intelligence application such as Google Assistant, Siri, Chatbots and so on. Many Deep learning algorithms such as LSTM, Bi-LSTM, RNN and CNN focus on generation of sentimental texts. The research work proposed herein uses Generative Adversarial Networks (GAN) for generating texts based on the training dataset. The GAN network has 2 models; generator model which generates the text and discriminator model which evaluates the text and both are trained using Bi-LSTM algorithm. The model is trained using Amazon customer review dataset and the system generates sentimental text such as positive, negative and neutral text for a given product. The existing system uses LSTM algorithm for prediction, but the performance and quality of generated text is low. The proposed system uses the Bi-LSTM algorithm which increases the accuracy of the generated sentimental text. The performance of the framework is evaluated in terms of fluency, sentimental accuracy and context relevance and is compared to the existing system of Automatic Generation of Sentimental Texts using Mixture Adversarial Networks.