An Augmented Neural Network for Sentiment Analysis Using Grammar.

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Human-Robot Interaction

Background:

• Deep learning models struggle with grammatical nuances in sentiment analysis.
• Accurate sentiment analysis is crucial for effective human-robot interaction.
• Existing models face challenges in representing linguistic variations.

Purpose of the Study:

• To propose a novel sentiment analysis model, MoLeSy, that incorporates morphological, lexical, and syntactic knowledge.
• To improve the accuracy of sentiment detection in natural language processing.
• To enhance human-robot interaction through better sentiment understanding.

Main Methods:

• Developed MoLeSy, a model augmenting neural networks with linguistic features.
• Utilized three concurrently processed neural networks: CNNs, LSTMs, and dense networks.
• Concatenated and reduced outputs from morphological, lexical, and syntactic channels using a dense layer.

Main Results:

• MoLeSy achieved superior performance compared to state-of-the-art models across four diverse sentiment analysis datasets (hotel, NLPCC2014, Douban, Weibo).
• Demonstrated the effectiveness of integrating morphological, lexical, and syntactic information.
• Showcased improved sentiment analysis accuracy.

Conclusions:

• Morphological, lexical, and syntactic grammar significantly augment neural networks for sentiment analysis.
• The MoLeSy model offers a promising advancement in natural language processing for sentiment detection.
• This research contributes to more sophisticated human-robot interaction capabilities.