Structure Graph Refined Information Propagate Network for Aspect-Based Sentiment Analysis

Introduction

Data mining sets a foundation of natural language processing. In our daily lives, people are constantly invited to share their opinions and preferences with the rest of the world, which results in an explosion of textual information. As such, data mining provides an opportunity to deal with the opinions on products, stocks, policies, and everything. In this context, sentiment analysis is thereby developed to determine the opinion of people regarding a given topic via textual data mining. Aspect-based sentiment analysis (ABSA) is currently an ongoing trend for precisely mining the user’s opinion.

Comprehensively, ABSA is a fine-grained sentiment analysis task in the field of sentiment analysis (Tang et al., 2014; Yang et al., 2017). The main purpose of ABSA is to identify the sentiment polarity (i.e., positive, neutral, or negative) toward a given aspect in a sentence or document. For instance, in the sentence ‘The service is decent even when this small place is packed,’ (Figure 1a), the two aspects ‘service’ and ‘place’ are extracted, whose sentiment polarities are classified as negative and positive, respectively. Aiming to deal with the issue of multiple aspects within one sentence, the interaction between an aspect and its contexts has to be resolved. The description of speech and dependency relationship is shown in Table 1.

Figure 1.

Two examples of syntax dependency tree