iACOS: Advancing Implicit Sentiment Extraction with Informative and Adaptive Negative Examples

TL;DR

The paper tackles implicit aspect and opinion quadruple extraction in ABSA by introducing iACOS, which appends implicit tokens $[IA]$ and $[IO]$ to capture hidden semantics, uses an extended BIOES scheme for joint explicit/implicit co-extraction, and employs a multi-label classifier with multi-head attention to predict category-sentiment pairs for each $(a,o)$ pair. It further introduces an informative and adaptive negative-sample strategy to train the model jointly across quadruple, category, and sentiment tasks within a multi-task learning framework, boosting data efficiency. Empirical results on two public datasets show strong F1 gains over state-of-the-art baselines, with ablation confirming the critical roles of implicit tokens, multi-head attention, and adaptive negatives. The method advances robust, end-to-end ABSA with implicit elements, offering practical impact for finer-grained sentiment extraction in real-world reviews.

Abstract

Aspect-based sentiment analysis (ABSA) have been extensively studied, but little light has been shed on the quadruple extraction consisting of four fundamental elements: aspects, categories, opinions and sentiments, especially with implicit aspects and opinions. In this paper, we propose a new method iACOS for extracting Implicit Aspects with Categories and Opinions with Sentiments. First, iACOS appends two implicit tokens at the end of a text to capture the context-aware representation of all tokens including implicit aspects and opinions. Second, iACOS develops a sequence labeling model over the context-aware token representation to co-extract explicit and implicit aspects and opinions. Third, iACOS devises a multi-label classifier with a specialized multi-head attention for discovering aspect-opinion pairs and predicting their categories and sentiments simultaneously. Fourth, iACOS leverages informative and adaptive negative examples to jointly train the multi-label classifier and the other two classifiers on categories and sentiments by multi-task learning. Finally, the experimental results show that iACOS significantly outperforms other quadruple extraction baselines according to the F1 score on two public benchmark datasets.

Figures (4)

• Figure 1: Framework of iACOS: left box for inference and right box for training with multi-tasking learning, in which negative sample construction is an important module.
• Figure 2: Convergence analysis on iACOS
• Figure 3: Effect of negative samples
• Figure 4: Ablation experiments