Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DINER: Debiasing Aspect-based Sentiment Analysis with Multi-variable Causal Inference

Jialong Wu, Linhai Zhang, Deyu Zhou, Guoqiang Xu

TL;DR

DINER tackles spurious correlations in ABSA by applying a multi-variable causal-inference framework that debiases both the review context and the target aspect. It uses backdoor adjustment to remove indirect confounding in the review branch and counterfactual reasoning to subtract direct aspect-to-label correlations, combining them into a debiased total indirect effect $TIE_{a,r}$. The model is trained with a multi-branch objective and evaluated on SemEval-derived data and the ARTS robustness benchmarks, achieving state-of-the-art results, especially with a SUM-$\tanh$ fusion strategy. The approach offers practical robustness gains for ABSA systems, though it assumes independence between the two causal inputs and is limited to two input variables in the current setup.

Abstract

Though notable progress has been made, neural-based aspect-based sentiment analysis (ABSA) models are prone to learn spurious correlations from annotation biases, resulting in poor robustness on adversarial data transformations. Among the debiasing solutions, causal inference-based methods have attracted much research attention, which can be mainly categorized into causal intervention methods and counterfactual reasoning methods. However, most of the present debiasing methods focus on single-variable causal inference, which is not suitable for ABSA with two input variables (the target aspect and the review). In this paper, we propose a novel framework based on multi-variable causal inference for debiasing ABSA. In this framework, different types of biases are tackled based on different causal intervention methods. For the review branch, the bias is modeled as indirect confounding from context, where backdoor adjustment intervention is employed for debiasing. For the aspect branch, the bias is described as a direct correlation with labels, where counterfactual reasoning is adopted for debiasing. Extensive experiments demonstrate the effectiveness of the proposed method compared to various baselines on the two widely used real-world aspect robustness test set datasets.

DINER: Debiasing Aspect-based Sentiment Analysis with Multi-variable Causal Inference

TL;DR

DINER tackles spurious correlations in ABSA by applying a multi-variable causal-inference framework that debiases both the review context and the target aspect. It uses backdoor adjustment to remove indirect confounding in the review branch and counterfactual reasoning to subtract direct aspect-to-label correlations, combining them into a debiased total indirect effect . The model is trained with a multi-branch objective and evaluated on SemEval-derived data and the ARTS robustness benchmarks, achieving state-of-the-art results, especially with a SUM- fusion strategy. The approach offers practical robustness gains for ABSA systems, though it assumes independence between the two causal inputs and is limited to two input variables in the current setup.

Abstract

Though notable progress has been made, neural-based aspect-based sentiment analysis (ABSA) models are prone to learn spurious correlations from annotation biases, resulting in poor robustness on adversarial data transformations. Among the debiasing solutions, causal inference-based methods have attracted much research attention, which can be mainly categorized into causal intervention methods and counterfactual reasoning methods. However, most of the present debiasing methods focus on single-variable causal inference, which is not suitable for ABSA with two input variables (the target aspect and the review). In this paper, we propose a novel framework based on multi-variable causal inference for debiasing ABSA. In this framework, different types of biases are tackled based on different causal intervention methods. For the review branch, the bias is modeled as indirect confounding from context, where backdoor adjustment intervention is employed for debiasing. For the aspect branch, the bias is described as a direct correlation with labels, where counterfactual reasoning is adopted for debiasing. Extensive experiments demonstrate the effectiveness of the proposed method compared to various baselines on the two widely used real-world aspect robustness test set datasets.
Paper Structure (24 sections, 14 equations, 3 figures, 8 tables)

This paper contains 24 sections, 14 equations, 3 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Aspect-Based Sentiment Analysis
  4. Causal Inference-based Debiasing
  5. Methods
  6. Structural Causal Model of ABSA
  7. Deconfounding the Review Branch with Backdoor Adjustment
  8. Decorrelating the Aspect Branch with Counterfactual Reasoning
  9. Training and Inference
  10. Experiments
  11. Datasets
  12. Baselines
  13. Implementations
  14. Evaluation
  15. Result and Analysis
  16. ...and 9 more sections

Figures (3)

  • Figure 1: (a) Examples are taken from the SemEval 2014 Restaurant test set. (b) RevTgt denotes reversing the polarity of the target aspect, RevNon denotes reversing the polarity of the non-target aspect, and AddDiff denotes adding another non-target aspect with different polarity.
  • Figure 2: (a) SCM of the proposed method. (b) The desired situation for ABSA, the dotted line means the causalities are blocked.
  • Figure 3: The framework of the proposed method.