From Graphs to Hypergraphs: Enhancing Aspect-Based Sentiment Analysis via Multi-Level Relational Modeling

TL;DR

ABSA faces conflicting sentiment signals and sparse context, especially in short texts. HyperABSA resolves this by constructing per-sentence dynamic hypergraphs through hierarchical clustering with an acceleration-fallback cutoff, enabling multi-node aspect–opinion modeling without external parsers or multi-graph fusion. The hypergraph is processed with HyperGAT and trained with cross-entropy plus regularization, achieving state-of-the-art performance on Lap14 and Rest14 with RoBERTa, and competitive results on MAMS, with extensive ablations confirming robustness and efficiency. This approach offers a principled, per-instance relational modeling paradigm that can extend to other short-text NLP tasks while reducing fusion complexity and error propagation.

Abstract

Aspect-Based Sentiment Analysis (ABSA) predicts sentiment polarity for specific aspect terms, a task made difficult by conflicting sentiments across aspects and the sparse context of short texts. Prior graph-based approaches model only pairwise dependencies, forcing them to construct multiple graphs for different relational views. These introduce redundancy, parameter overhead, and error propagation during fusion, limiting robustness in short-text, low-resource settings. We present HyperABSA, a dynamic hypergraph framework that induces aspect-opinion structures through sample-specific hierarchical clustering. To construct these hyperedges, we introduce a novel acceleration-fallback cutoff for hierarchical clustering, which adaptively determines the level of granularity. Experiments on three benchmarks (Lap14, Rest14, MAMS) show consistent improvements over strong graph baselines, with substantial gains when paired with RoBERTa backbones. These results position dynamic hypergraph construction as an efficient, powerful alternative for ABSA, with potential extensions to other short-text NLP tasks.

Figures (6)

• Figure 1: A hypergraph of word interactions showing several semantic clusters based on aspect and sentiment polarity. This illustrates how words are grouped according to meaning and sentiment.
• Figure 2: Architecture of HyperABSA.
• Figure 3: Dendrogram visualization of token-level hierarchical clustering.
• Figure 4: Evaluation of HyperABSA against multi-graph-based models on the Rest14 and MAMS datasets in terms of generalization gap.
• Figure 5: Visualization of the hypergraph structure for the representative input, illustrating the model’s ability to form semantically coherent token groupings.