Authors:
Uroš Hercog, Andraž Povše
Abstract:
Determining the trustworthiness of individual Bitcoin wallets remains a significant challenge. Currently, there are no standardized ratings that provide meaningful information about the taint level—how "dirty" or potentially illicit—the received Bitcoin might be. This lack of transparency exposes transactions to liability risks, especially if the Bitcoin originates from theft or illegal activities.
In this study, we introduce TaintRank, a novel scoring system designed to evaluate the taint level of Bitcoin addresses. By analyzing the entire transaction history associated with a wallet and its interactions with other addresses, TaintRank offers deep insights into wallet credibility. This ranking method empowers Bitcoin exchange companies with actionable intelligence about their transaction counterparts.
Predicting the Topical Stance of Media and Popular Twitter Users
Authors:
Peter Stefanov, Kareem Darwish, Preslav Nakov
Abstract:
Controversial socio-political issues drive intense online discourse, with users frequently sharing media articles and amplifying statements from influential figures. Understanding the stance of both media outlets and prominent Twitter users on these topics is crucial for policymakers and researchers.
While supervised solutions for stance detection exist, manual annotation is costly. This paper proposes an unsupervised learning approach that leverages retweet behavior to characterize political leanings and topical stances. Our model’s predictions are validated against gold-standard labels from media bias/fact-checking platforms, supplemented by human analysis.
Danish Stance Classification and Rumour Resolution
Authors:
Anders Edelbo Lillie, Emil Refsgaard Middelboe
Abstract:
The internet is rife with rumors propagated via blogs and social media. Recent studies suggest that analyzing crowd stance toward rumors serves as a reliable indicator of their veracity.
This study generates an annotated Reddit dataset for Danish and implements various stance classification models. A linear SVM achieves the best performance (76% accuracy). Additionally, experiments demonstrate cross-language applicability, using stance labels with Hidden Markov Models (HMMs) to predict rumor accuracy (83% accuracy for Danish data).
Key Insights
- Bitcoin Transparency: TaintRank addresses a critical gap in cryptocurrency accountability by quantifying wallet trustworthiness.
- Media Bias Detection: Unsupervised retweet-based analysis effectively maps political stances without labeled data.
- Rumor Debunking: Stance classification combined with HMMs offers a scalable solution for multilingual rumor verification.
👉 Explore more about blockchain analytics
FAQ
Q: How does TaintRank differ from existing Bitcoin analysis tools?
A: TaintRank evaluates historical interactions holistically, unlike tools focusing solely on direct transactions.
Q: Can the stance detection model be applied to other languages?
A: Yes, the framework is language-agnostic, though performance varies based on data availability.
Q: What practical applications does rumor resolution have?
A: It aids platforms in flagging misinformation and helps users discern credible sources.
👉 Learn how SEO optimizes financial content