A Method of Sentiment Classification for Chinese Aspect Level Based on Information Fusion
Changyuan Wu,Dancheng Li,Ouwen Zhu
DOI: https://doi.org/10.1109/icaibd57115.2023.10206290
2023-01-01
Abstract:In this paper, we present an enhanced approach to Aspect-Based Sentiment Classification (ABSC), a vital subtask of Aspect-Based Sentiment Analysis (ABSA) that focuses on identifying the sentiment polarity (positive, neutral, or negative) towards specific aspects within textual data, such as product reviews or customer feedback. This fine-grained analysis holds significant value for businesses and organizations, as it enables them to gain deeper insights into customer opinions and preferences. Although recent research has shown that fine-tuning pre-trained models (PTMs) is more effective for addressing ABSC challenges compared to previous syntax-based methods, there remains room for improvement. We propose a novel model that fine-tunes PTMs by concurrently considering context information, opinion words information, and prompt information, effectively capturing the intricate relationships between aspects and their corresponding sentiment expressions. This comprehensive approach allows for a more accurate classification of sentiment polarity for each aspect. We evaluate our model on several Chinese datasets, covering various domains such as hotels, cars, and phones, demonstrating its robustness and adaptability. On these Chinese datasets, our proposed model outperforms the previous state-of-the-art (SOTA) models in terms of accuracy. Additionally, we present an in-depth analysis of our model’s performance, discussing its strengths and limitations, and provide suggestions for future research directions in the field of ABSC.