Aspect-level sentiment analysis incorporating multidimensional feature

Aspect-Based Sentiment Analysis (ABSA) belongs to the task of fine-grained sentiment analysis, aiming to understand people’s sentiment polarity towards evaluation targets at the aspect level. In recent years, significant progress has been made in related research, but the existing methods focus on utilizing the semantic and syntactic information of sentences, without making full use of the textual thematic features as well as the importance level of aspect words. In addition, existing graph-based neural network models have insufficient ability to node feature retention. To address this problem, firstly, based on the syntactic dependency tree, the dependency types between context words and aspect words are fully explored and integrated into the construction of the dependency graph; secondly, a global graph attention network is used to construct the syntactic dependency tree; lastly, the introduction of textual topic features makes the model Finally, the introduction of text topic features makes the model pay more attention to topic-related emotion words. The proposed ITRGAT model integrates the syntactic and contextual semantic representations learned in parallel to accomplish sentiment enhancement and syntactic enhancement. Extensive experiments are conducted on three public datasets to demonstrate the effectiveness of ITRGAT.