Incorporating Syntactic Knowledge and Position Information for Aspect-Based Sentiment Analysis

Aspect-based sentiment analysis (ABSA) is a subtask of fine-grained sentiment analysis that focuses on analysing the sentiment polarity of aspect terms in a given sentence. Previous studies have extracted sentiment interaction information between aspect and target sentiment words with a dependency syntax parse tree (DSPT). However, those models relied excessively on DSPT and thus did not perform well in identifying local context information. Moreover, modelling syntactic information equivalently with complex dependency information may introduce noise and degrade the model’s performance. Therefore, we incorporate sentence constituent information into DSPT such that the model can learn the association information not only within a sentence but also between remote words. Furthermore, we also capture the sentiment interaction information between aspect terms based on their distance in the context to reduce internal noise and the effect of irrelevant words.