MiMuSA: Mimicking human language understanding for fine-grained multi-class sentiment analysis

• Main Authors: WANG, Zhaoxia, HU, Zhenda, HO, Seng-Beng, CAMBRIA, Erik, TAN, Ah-hwee
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2023
• Subjects: Human-like understanding; Fine-grained sentiment understanding; Multi-class sentiment analysis; Sentiment strength; Explainable sentiment understanding; Sarcasm handling; Knowledge base; Multi-level modularstructure; Artificial Intelligence and Robotics; Numerical Analysis and Scientific Computing
• Online Access: https://ink.library.smu.edu.sg/sis_research/7953; https://ink.library.smu.edu.sg/context/sis_research/article/8956/viewcontent/MiMuSA_av.pdf
• Institution: Singapore Management University

Sentiment analysis is an important natural language processing (NLP) task due to a wide range of applications. Most existing sentiment analysis techniques are limited to the analysis carried out at the aggregate level, merely providing negative, neutral and positive sentiments. The latest deep learning-based methods have been leveraged to provide more than three sentiment classes. However, such learning-based methods are still black-box-based methods rather than explainable language processing methods. To address this gap, this paper proposes a new explainable fine-grained multi-class sentiment analysis method, namely MiMuSA, which mimics the human language understanding processes. The proposed method involves a multi-level modular structure designed to mimic human’s language understanding processes, e.g., ambivalence handling process, sentiment strength handling process, etc. Specifically, multiple knowledge bases including Basic Knowledge Base, Negation and Special Knowledge Base, Sarcasm Rule and Adversative Knowledge Base, and Sentiment Strength Knowledge Base are built to support the sentiment understanding process. Compared with other multi-class sentiment analysis methods, this method not only identifies positive or negative sentiments, but can also understand fine-grained multi-class sentiments, such as the degree of positivity (e.g., strongly positive or slightly positive) and the degree of negativity (e.g., slightly negative or strongly negative) of the sentiments involved. The experimental results demonstrate that the proposed MiMuSA outperforms other existing multi-class sentiment analysis methods in terms of accuracy and F1-Score.