it may had performed well on test data if i had used more epochs while training, used Recurrent Neural Neural or stack of Recurrent Neural Networks with dropouts and regularization.
there is a lot of gap between training and validations loss and accuracy, this is because i used pretrained word embedding GLOVE which knows the relations between sentiments.
this is performed on a normal pc, no gpu has been used, therefore you can analyse the amount of time i had to spent while visualizing my approach and efficiency.