Implicit aspect-based generative model for sentiment analysis based on prompt learning
Xinlong Wang, Xu Li, Yinghui Yin, Yang Li · 2024
In aspect-based sentiment analysis, there may be implicit aspect items and opinion items in the text that do not appear explicitly but can be identified through semantic reasoning. The existence of this implicit information increases the complexity and challenge of semantic understanding. Since these terms do not appear directly in the text, the model needs to have a deeper understanding of the context and capture hidden semantic clues to correctly identify and analyze these implicit aspects and sentiment relations. In order to solve the problem of implicit aspect-based sentiment analysis, this paper proposes a generative model for implicit aspect-based sentiment analysis based on prompt learning, which employs a generative pre-training model T5 and combines it with prompt learning to capture prompt and semantic information in the context to understand implicit sentiments, as well as the aspects and opinions in the quadruple task are expressed in the form of an index, which is intended to reduce the dimensionality of the output space to improve the accuracy of the model. Compared to the existing state-of-the-art models, the proposed model improves the F1 value by $\mathbf{0 . 4 7 \%}$ on the standard dataset restaurantACOS and $0.64 \%$ on the LaptopACOS dataset. Ablation experiments demonstrated that each improvement was effective for aspect-based sentiment analysis.