MuCon: Multi-channel convolution for targeted sentiment classification

Targeted Sentiment Analysis goes beyond general sentiment classification tasks by aiming to identify the sentiment of a specific target aspect within a given text. Previous studies have predominantly utilized recurrent neural networks (RNN) or their variants to predict target-specific sentiment polarity. However, the sequential processing nature of RNN restricts parallelization and fails to leverage the potential of modern multicore architectures. Additionally, these models often overlook the inherent linguistic perspective embedded in the text. This paper proposes a novel approach called MuCon (Multi-channel Convolution), which employs a simple yet effective convolutional neural network (CNN) model. MuCon incorporates multiple channels dedicated to linguistic and statistical features to determine aspect-specific sentiment polarity accurately. By incorporating linguistic knowledge into a statistical model, MuCon performs better and achieves comparable results to sophisticated state-of-the-art methods.

Automated summary

Targeted Sentiment Analysis aims to identify the sentiment of a specific target aspect in a given text, going beyond general sentiment classification tasks. Previous studies mainly use recurrent neural networks (RNN) or variants to predict target-specific sentiment polarity, but RNN's sequential processing nature limits parallelization and fails to fully leverage modern multicore architectures' potential. Additionally, these models often overlook the inherent linguistic perspective embedded in the text. This paper proposes a novel approach called MuCon (Multi-channel Convolution), which uses a simple yet effective convolutional neural network (CNN) model. MuCon accurately determines aspect-specific sentiment polarity by incorporating multiple channels dedicated to linguistic and statistical features. By incorporating linguistic knowledge into a statistical model, MuCon performs better and achieves comparable results to sophisticated state-of-the-art methods.