Recurrent Convolutional Attention Neural Model for Sentiment Classification of short text

Nowadays neural attention model doing good in many tasks of natural language processing (NLP). More specifically attention mechanism has been used widely with convolutional neural network (CNN) and recurrent neural network (RNN) for many task of NLP. But the sentiment classification of short text is a challenging task because it contains limited contextual information. Thus, we proposed a new recurrent convolutional attention neural model for sentiment classification of the short text by using attention mechanism with RCNN (recurrent convolutional neural network). In proposed model attention score is calculated by averaging hidden units (feature maps) generated from LSTM (long short-term memory). Then we combined this attention score with recurrent convolution-based encoded text features to get final sentence representation. Here attention will focus on important text features and recurrent convolution makes full use of limited contextual information by processing sentence representation through different windows sizes with specialized recurrent convolution operation. Validation of the proposed model is done through experimentation with three benchmark datasets i.e. MR, SSTl, and SST2. Achieved results exhibit that our model performs better than many existing baselines works on all three datasets.