Single-Pass, Depth-Selective Reading for Multi-Aspect Sentiment Analysis

TL;DR

DABS introduces a single-pass, depth-selective reading method for multi-aspect sentiment analysis, significantly reducing computation while maintaining competitive accuracy by adaptively querying shared sentence representations.

Contribution

It proposes a novel framework that decouples sentence encoding from aspect-specific reading, enabling efficient, adaptive, multi-aspect sentiment analysis without re-encoding.

Findings

Achieves up to 60% reduction in computation in multi-aspect scenarios.

Maintains competitive performance across four ATSA benchmarks.

Most benefits observed in linguistically complex cases like negation and contrast.

Abstract

Aspect-Term Sentiment Analysis (ATSA) in multi-aspect sentences faces a fundamental tradeoff between efficiency and expressiveness. Existing models either re-encode the sentence for each aspect or rely on static use of deep representations, leading to redundant computation and limited adaptivity. We argue that Transformer depth is a costly, queryable resource, and propose DABS, a single-pass inference framework that encodes each sentence once to construct a reusable, depth-ordered substrate. Each aspect then queries this shared representation to selectively read relevant tokens and abstraction levels, without re-encoding. This decouples shared sentence encoding from lightweight, aspect-conditioned readout. Experiments on four ATSA benchmarks show that DABS achieves competitive performance while reducing end-to-end computation by up to 60% in multi-aspect settings (M >= 2). Further analyses indicate that adaptive depth querying is most beneficial for linguistically complex cases such as negation and contrast. Code is publicly available at https://github.com/panzhzh/acl-dabs