On the Robustness of Aspect-based Sentiment Analysis: Rethinking Model, Data, and Training

Abstract

Aspect-based sentiment analysis (ABSA) aims at automatically inferring the specific sentiment polarities towards certain aspects of products or services behind the social media texts or reviews, which has been a fundamental application to the real-world society. Within recent decade, ABSA has achieved extraordinarily high accuracy with various deep neural models. However, existing ABSA models with strong in-house performances may fail to generalize to some challenging cases where the contexts are variable, i.e., being low robustness to real-world environment. In this study, we propose to enhance the ABSA robustness by systematically rethinking the bottlenecks from all possible angles, including model, data and training. First, we strengthen the current best-robust syntax-aware models by further incorporating the rich external syntactic dependencies and the labels with aspect simultaneously with a universal-syntax graph convolutional network. In the corpus perspective, we propose to automatically induce high-quality synthetic training data with various types, allowing models to learn sufficient inductive bias for better robustness. Lastly, we based on the rich pseudo data perform adversarial training to enhance the resistance to the context perturbation, and meanwhile employ contrastive learning to reinforce the representations of instances with contrastive sentiments. Extensive robustness evaluations are conducted. The results demonstrate that our enhanced syntax-aware model achieves better robustness performances than all the state-of-the-art baselines. By additionally incorporating our synthetic corpus, the robust testing results are pushed with around 10% accuracy, which are then further improved by installing the advanced training strategies. In-depth analyses are presented for revealing the factors influencing the ABSA robustness.

User Opinion Prediction for Arabic Hotel Reviews Using Lexicons and Artificial Intelligence Techniques

“…However, observations have shown that aspect-based sentiment analysis models trained on a dataset from one domain do not generalize well to another dataset belonging to another domain [23]. In [24], the authors enhanced the aspect-based sentiment analysis robustness by improving the model, data, and training. They tested data from different domains.…”

Position-Wise Gated Res2Net-Based Convolutional Network with Selective Fusing for Sentiment Analysis

“…Graph convolutional networks (GCNs), which are an adaptation of the CNNs for handling unstructured data, can facilitate the handling of dependency trees. Zhang et al [22] built a universal-syntax GCN over the syntactic dependencies with labels to achieve the goal of navigating richer syntax information for the best aspect-based SA robustness. With the development of deep learning, several strategies and approaches have been proposed for improving the ability of CNNs to extract sentiment features [23].…”

Application of SVM and Chi-Square Feature Selection for Sentiment Analysis of Indonesia’s National Health Insurance Mobile Application

“…The margin 𝛾 can be obtained by calculating the length of the projection of the difference between the support vectors and the 𝐰 vector in Equation (12).…”

“…Shi et al (2022) addressed limitations in structured sentiment analysis by proposing a novel labeling strategy and a graph attention network-based model, significantly surpassing previous state-of-the-art models [11]. Fei et al (2022) focused on enhancing the robustness of ABSA models through multi-faceted improvements, spanning model design, data augmentation, and advanced training strategies [12].…”

“…Moreover, Huang et al (2020) introduced a weakly supervised approach for aspectbased sentiment analysis, utilizing sentiment aspect joint topic embeddings and neural classifiers to overcome the absence of labeled examples [13]. Li et al (2022) bridged the gap between sentiment analysis and dialogue contexts by introducing the conversational aspect-based sentiment quadruple analysis task, and providing a benchmark dataset and model for cross-utterance quadruple extraction [14]. Another contribution by Fei (2020) involved the development of a Latent Emotion Memory network for multi-label emotion classification, integrating latent emotion distribution and context information to achieve state-of-the-art results [15].…”