What is it about?
This paper reviews research on implicit aspect extraction—the task of identifying what exactly someone is commenting on when they don’t say it directly. For instance, a sentence can express an opinion without naming the target feature, so the aspect must be inferred from context. The paper explains why these “hidden” targets matter for aspect-based sentiment analysis and surveys how researchers have approached the problem so far. It systematically summarizes commonly used datasets, algorithmic/modeling approaches, and the end-to-end pipeline used in prior work, while also discussing the challenges that have slowed progress in this area.
Featured Image
Photo by Amanz on Unsplash
Why is it important?
Implicit aspects (targets of user opinions) are not rare—they can make up more than 20% of aspects in a dataset, and missing them can hurt the accuracy of aspect-based analyses. A substantial portion of opinion targets in widely used benchmarks are implicit (e.g., ~25% in SemEval 2015/2016 opinion targets; and high implicit rates reported in Laptop-ACOS and Restaurant-ACOS). Because of this, implicit aspect extraction affects the performance of downstream systems that try to extract aspects/categories, opinions, and sentiment—especially in real-world feedback where people often imply what they mean rather than naming it explicitly. The paper is also important as a field resource: it follows a PRISMA-style systematic process and reports how the literature was filtered down to the final set of included studies, providing a clear map of what has been done and what gaps remain.
Perspectives
One thing I wanted this review to accomplish was to make the implicit aspect extraction landscape easier to navigate for both newcomers and active researchers. Following a PRISMA-style search and screening process, we started from a large pool of papers and ultimately selected 48 articles for inclusion, covering research published from 2015 to 2022—with the goal of being comprehensive and up to date. Writing this also made the key bottlenecks very clear: the field is held back by inconsistent definitions (what counts as an aspect vs. a feature vs. an aspect category), limited benchmark/gold datasets, and historically limited use of stronger neural/attention-based methods—though newer transformer and generative models are starting to appear. Overall, I hope this paper serves as a practical reference point and encourages clearer problem definitions, more shareable datasets, and stronger evaluations so progress becomes easier to measure and compare.
Meghna Chaudhary
University of South Florida
Read the Original
This page is a summary of: Implicit Aspect Extraction: A Systematic Review, ACM Computing Surveys, December 2025, ACM (Association for Computing Machinery),
DOI: 10.1145/3786590.
You can read the full text:
Contributors
The following have contributed to this page
