Text Generation Model Enhanced with Semantic Information in Aspect Category Sentiment Analysis

Aspect Category Sentiment Analysis (ACSA) is one of the main subtasks of sentiment analysis, which aims at predicting polarity over a given aspect category. Recently, generative methods have emerged as an efficient way to use a pre-trained language model for ACSA. However, those methods fail to model relations between target words and opinion words in a sentence including multiple aspects. To tackle this problem, this paper proposes a method to incorporate Abstract Meaning Representation (AMR), which describes the semantic representation of a sentence as a directed graph, into a text generation model. Furthermore, two regularizers are designed to guide the allocation of cross attention weights over AMR graphs. One is the identical regularizer, which constrains the attention weights of aligned nodes, the other is the entropy regularizer, which helps the decoder generate tokens by only giving a high degree of consideration to a few related nodes in the AMR graph. Experimental results on three datasets show that the proposed method outperforms state-of-the-art methods, proving the effectiveness of our model.