Gated recurrent neural network with sentimental relations for sentiment classification

Gated recurrent neural networks (GRNNs) have been very successful in sentiment classification due to their ability to preserve semantics over time. However, modeling sentimental relations such as negation and intensification under a recurrent architecture remains a challenge. In this work, we introduce a gated recurrent neural network with sentimental relations (GRNN-SR)(1) to capture the sentimental relations' information from sentiment modifier context and model their effects in texts. At each time step, GRNN-SR separately encodes the information of sentiment polarity and sentiment modifier context. The new sentiment inputs are modified multiplicatively by the previous encoded sentiment modifier context before they are updated into current states of sentiment polarity, which is more effective than the approach of traditional GRNNs. The experimental results show that our model not only can capture sentimental relations but also is an improvement over state-of-the-art gated recurrent neural network baselines. (C) 2019 Elsevier Inc. All rights reserved.