Papers › Masking The Bias : From Echo Chambers to Large Scale Aspect-Based Sentiment Analysis

Masking The Bias : From Echo Chambers to Large Scale Aspect-Based Sentiment Analysis

2 Sep 2024IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM) 2024 9archive 2025-07-28

Yeonjung Lee, Yusuf Mücahit Çetinkaya, Emre Külah, İsmail Hakkı Toroslu, Hasan Davulcu

Aspect-based sentiment analysis (ABSA) is a natural language processing (NLP) task, ascribing precise sentiment linkages to specific entities and issues in text data. This paper addresses critical shortcomings in current ABSA methods, particularly the issues of limited aspects, training set biases, and lack of comprehensive stance-coded datasets. First, we develop a scalable MaskedABSA approach that masks aspect terms in training sentences to enable unbiased sentiment inference from the context alone. We show that the proposed method surpasses the state-of-the-art solutions in accuracy for the aspect term sentiment classification task, as verified by the SemEval datasets. Furthermore, we tackle the perennial challenges of limited training resources and the prohibitive costs of manual annotation in ABSA dataset creation by introducing an innovative weak supervision technique capitalizing on the inherent community clustering properties found within social media datasets. We utilize community detection algorithms to partition a share network into polarized groups with homogeneous adversarial stances, allowing large-scale aspect-based sentiment analysis dataset curation without labor intensive manual labeling. Our methodology is also validated using a real-world polarized dataset comprising diverse aspects and stances to showcase its efficacy and scalability.

Code

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Results from the paper archive 2025-07-28

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections