Target-Based Attention Model for Aspect-Level Sentiment Analysis

Aspect-level sentiment classification, which aims to determine the sentiment polarity of the specific target word or phrase of a sentence, is a crucial task in natural language processing (NLP). Previous works have proposed various attention methods to capture the important part of the context for the desired target. However, these methods have less interaction between aspects and contexts and can not accurately quantify the importance of context words with the information of aspect. To address these issues, we firstly proposed a novel target-based attention model (TBAM) for aspect-level sentiment analysis, which employs an attention mechanism between the position-aware context representation matrix. TBAM can generate more accurate attention scores between aspects and contexts at the word level in a joint way, and generate more discriminative features for classification. Experimental results show that our model achieves a state-of-the-art performance on three public datasets compared to other architectures.