Domain Generalization via Causal Adjustment for Cross-Domain Sentiment Analysis

TL;DR

This work addresses domain generalization in cross-domain sentiment analysis by recasting the task within a structural causal model and using backdoor adjustment to disentangle domain-invariant and domain-specific representations. The authors design a training-time invariant learning framework that enforces a Backdoor Condition, combining a domain-invariant encoder with a domain-specific branch and a set of losses to align $P(Y\mid M_{inv})$ with $P(Y\mid do(M_{inv}))$. Empirical results across more than 20 homologous and diverse datasets show strong generalization to unseen domains, with ablations confirming the importance of both invariant and domain-specific components, and representation visualizations illustrating improved invariance. The approach also compares with large language models, revealing domain generalization challenges for LLMs and highlighting practical gains for targeted DG in sentiment analysis.

Abstract

Domain adaption has been widely adapted for cross-domain sentiment analysis to transfer knowledge from the source domain to the target domain. Whereas, most methods are proposed under the assumption that the target (test) domain is known, making them fail to generalize well on unknown test data that is not always available in practice. In this paper, we focus on the problem of domain generalization for cross-domain sentiment analysis. Specifically, we propose a backdoor adjustment-based causal model to disentangle the domain-specific and domain-invariant representations that play essential roles in tackling domain shift. First, we rethink the cross-domain sentiment analysis task in a causal view to model the causal-and-effect relationships among different variables. Then, to learn an invariant feature representation, we remove the effect of domain confounders (e.g., domain knowledge) using the backdoor adjustment. A series of experiments over many homologous and diverse datasets show the great performance and robustness of our model by comparing it with the state-of-the-art domain generalization baselines.

Figures (5)

• Figure 1: Difference between domain adaptation and domain generalization.
• Figure 2: Structural Causal Model of Cross-domain Sentiment analysis
• Figure 3: The framework of our model.
• Figure 4: Results on 13 Homologous datasets.
• Figure 5: Representation visualization.