Stance classification

Stance classification is the task of determining whether a piece of text (e.g., a tweet, comment, or statement) supports, opposes, questions, or comments on a given target—which may be a claim, entity, or proposition.

Applications

• Rumour verification: Analyzing how users respond to an alleged claim in a conversation thread
• Claim verification: Understanding community consensus on a factual claim
• Argumentation mining: Identifying argument relationships in debates
• Sentiment-aware analysis: Distinguishing opinion direction from stance

Key distinction: stance vs. sentiment

• Sentiment: Subjective emotional tone (positive/negative/neutral)
• Stance: Position toward a target or proposition (support/oppose/neutral/query/comment)

A tweet may express negative sentiment but support a claim, or vice versa.

Related work

Key papers

• Kumar & Carley (2019) — Tree LSTMs with Convolution Units to Predict Stance and Rumor Veracity in Social Media Conversations — proposes Tree LSTM with convolutional child aggregation units for conversation-aware stance classification; achieves 0.520 mean F1-macro on PHEME across five events, outperforming prior work by 12%; demonstrates that convolutional operations learning source-reply contrasts outperform sum/max-pooling aggregation.
• Zubiaga et al. (2016) — Stance Classification in Rumours as a Sequential Task Exploiting the Tree Structure of Social Media Conversations — introduces Linear and Tree CRF variants to exploit sequential and tree-structured conversational patterns; demonstrates benefit of tree-structured representation
• Kochkina et al. (2017) — Sequential Approach to Rumour Stance Classification with Branch-LSTM — state-of-the-art neural approach for conversation-aware stance classification, achieving 78.4% accuracy on RumourEval 2017
• SemEval-2017 Task 8: RumourEval — SDQC (Support/Deny/Query/Comment) framework for stance in rumour conversations
• Zubiaga et al. (2018) — Detection and Resolution of Rumours in Social Media: A Survey — comprehensive review of stance classification approaches in rumour verification, annotation schemes (SDQC), and datasets
• Kochkina et al. (2018) — All-in-one: Multi-task Learning for Rumour Verification — demonstrates stance classification as effective auxiliary task for improving veracity prediction via multi-task learning