A Systematic Review of Aspect-based Sentiment Analysis: Domains, Methods, and Trends

TL;DR

The paper conducts a large-scale systematic literature review of ABSA across 2008-2024. It analyzes 519 in-scope ABSA studies (with a Phase-2 addition of 208 Gen-LLM ABSA studies, totaling 727 primary studies) to uncover domain and dataset skews, subtask distributions, and evolving solution paradigms. Key findings reveal a strong bias toward product/service datasets and SemEval benchmarks, with public-domain datasets underrepresented, and a heavy reliance on labeled data. The authors discuss implications for future research, emphasizing domain adaptation, diverse datasets, data-efficient approaches, and cautious integration of foundation models and in-context learning to broaden ABSA applicability and robustness.

Abstract

Aspect-based sentiment analysis (ABSA) is a fine-grained type of sentiment analysis that identifies aspects and their associated opinions from a given text. With the surge of digital opinionated text data, ABSA gained increasing popularity for its ability to mine more detailed and targeted insights. Many review papers on ABSA subtasks and solution methodologies exist, however, few focus on trends over time or systemic issues relating to research application domains, datasets, and solution approaches. To fill the gap, this paper presents a systematic literature review (SLR) of ABSA studies with a focus on trends and high-level relationships among these fundamental components. This review is one of the largest SLRs on ABSA. To our knowledge, it is also the first to systematically examine the interrelations among ABSA research and data distribution across domains, as well as trends in solution paradigms and approaches. Our sample includes 727 primary studies screened from 8550 search results without time constraints via an innovative automatic filtering process. Our quantitative analysis not only identifies trends in nearly two decades of ABSA research development but also unveils a systemic lack of dataset and domain diversity as well as domain mismatch that may hinder the development of future ABSA research. We discuss these findings and their implications and propose suggestions for future research.

Figures (10)

• Figure 1: Number of Studies by Publication Year: Total reviewed (N=4191) vs. Included (N=519)
• Figure 2: Distribution of Unique "Study-Dataset" Pairs (N=1179, with 519 studies and 218 datasets) by Research (Application) Domains (left) and Dataset Domains (right). Note: 1) The top flow visualises a mismatch between the two domains: the majority of studies without a specific research domain used datasets from the product/service review domain. 2) The disproportionately small number of samples in both domains that were neither "non-specific" nor "product/service review".
• Figure 3: Number of In-Scope Studies by Research (Application) Domain and Publication Year (N=518). This graph excludes the one 2023 study (extracted in October 2022) to avoid trend confusion.
• Figure 4: Number of Studies Per Each Research (Application) Domain (left), Dataset Domain (middle), and Dataset (right) Combination, filtered by Datasets Used by 10 or More In-Scope Studies (N=757). The three-way relationship highlights that not only did the majority of the sample studies with "non-specific" research domain use datasets from the 'product/service review' domain, but their datasets were also dominated by only four SemEval datasets on two types of product and service reviews.
• Figure 5: Number of Studies by ABSA Subtask