The Constrained Interaction Network for Aspect-level Sentiment Classification Task

Abstract

The purpose of aspect-level sentiment classification is to predict the sentiment polarity of specific aspect words in a sentence. Recently many works exploit LSTM models based on the attention mechanism. However, the prior work only attends to using the aspect terms to capture the aspect-specific sentiment information in the text. It may cause the mismatch of sentiment when the aspect words are extracted incorrectly. To solve this problem, we propose a simple but effective framework called the Constrained Interaction Network(CIN), which consists of the context-aspect level interaction layer(CAI-Layer), the long and short-term memory network layer(LSTM-Layer), and Constraint Attention layer(CA-Layer). CIN can extract the sentiment features of specific aspects with the assistance of LSTM-Layer and CAI-Layer, which greatly share the attention layer. The experiment conducted on three widely used data sets in SemEval 2014 and Twitter shows that the constrained attention mechanism is always better than other existing attention mechanisms, which also confirms that the CA- Layer can indeed help LSTM to extract the specified aspect-level sentiment characteristics.