CoSD: Collaborative Stance Detection with Contrastive Heterogeneous Topic Graph Learning

TL;DR

CoSD addresses shortcomings of fully parametric stance detectors by introducing contrastive heterogeneous topic graph learning that links texts, implicit topics, and stances. It constructs a Heterogeneous Topic Graph using LDA-derived implicit topics, and employs a Collaboration Propagation Aggregation module to capture multi-hop collaborative signals, refined through a contrastive training objective. A hybrid inference stage combines semantic signals from BERT with distributional signals from implicit topics to achieve robust stance predictions, achieving state-of-the-art results on SemEval-2016 and UKP while enhancing explainability through topic-aware reasoning. The approach demonstrates strong performance and interpretability, offering a scalable framework for leveraging latent topic structure in stance detection with potential for zero-shot extensions.

Abstract

Stance detection seeks to identify the viewpoints of individuals either in favor or against a given target or a controversial topic. Current advanced neural models for stance detection typically employ fully parametric softmax classifiers. However, these methods suffer from several limitations, including lack of explainability, insensitivity to the latent data structure, and unimodality, which greatly restrict their performance and applications. To address these challenges, we present a novel collaborative stance detection framework called (CoSD) which leverages contrastive heterogeneous topic graph learning to learn topic-aware semantics and collaborative signals among texts, topics, and stance labels for enhancing stance detection. During training, we construct a heterogeneous graph to structurally organize texts and stances through implicit topics via employing latent Dirichlet allocation. We then perform contrastive graph learning to learn heterogeneous node representations, aggregating informative multi-hop collaborative signals via an elaborate Collaboration Propagation Aggregation (CPA) module. During inference, we introduce a hybrid similarity scoring module to enable the comprehensive incorporation of topic-aware semantics and collaborative signals for stance detection. Extensive experiments on two benchmark datasets demonstrate the state-of-the-art detection performance of CoSD, verifying the effectiveness and explainability of our collaborative framework.

Figures (8)

• Figure 1: An illustrative example demonstrates the impact of collaboration on the test text $T$ and how they contribute to producing the correct result. In the example, the thick brown arrow signifies the generation of implicit topics from the training set. The thin black arrows symbolize the discovery of implicit topics within the test text. The texts $T_1$, $T_i$, and $T_n$ are similar to $T$ in terms of their implicit topic ($IT$) distribution, and they aid $T$ in making accurate judgments, specifically in favor of the target "abortion" in the UKP dataset.
• Figure 2: The overall framework structure of our CoSD stance detection method.
• Figure 3: Performance of CoSD under different values of the parameter topic numbers $H$
• Figure 4: Performance of CoSD under different values of the parameter hop numbers $l$
• Figure 5: T-SNE visualization of Target 'HC' on SemEval-2016