Test-time scaling has enabled Large Language Models (LLMs) to tackle complex reasoning, yet the limitations of current Chain-of-Thought (CoT) evaluation obscure whether performance gains stem from genuine reasoning or mere verbosity.To address this, (1) we propose a novel neuro-symbolic framework for the nonintrusive, comprehensive process-centric evaluation of reasoning grounded in First-Order Logic.(2) Through this lens, we identify four distinct behavioral prototypes and diagnose the failure modes.(3) We examine the impact of inference mode, training strategy, and model scale.Our analysis reveals that extended token generation is not a prerequisite for deep reasoning.Furthermore, we reveal critical constraints: mixing long and short CoT data in training risks premature saturation and collapse, while distillation into smaller models captures behavioral length but fails to replicate logical efficacy due to intrinsic capacity limits.

Group conversations over videoconferencing are a complex social behavior. However, the subjective moments of negative experience, where the conversation loses fluidity or enjoyment remain understudied. These moments are infrequent in naturalistic data, and thus training a supervised learning (SL) model requires costly manual data annotation. We applied semi-supervised learning (SSL) to leverage targeted labeled and unlabeled clips for training multimodal (audio, facial, text) deep features to predict non-fluid or unenjoyable moments in holdout videoconference sessions. The modality-fused co-training SSL achieved an ROC-AUC of 0.9 and an F1 score of 0.6, outperforming SL models by up to 4% with the same amount of labeled data. Remarkably, the best SSL model with just 8% labeled data matched 96% of the SL model's full-data performance. This shows an annotation-efficient framework for modeling videoconference experience.

Blockchain technology has the characteristics of decentralization, traceability and tamper-proof, which creates a reliable decentralized trust mechanism, further accelerating the development of blockchain finance. However, the anonymization of blockchain hinders market regulation, resulting in increasing illegal activities such as money laundering, gambling and phishing fraud on blockchain financial platforms. Thus, financial security has become a top priority in the blockchain ecosystem, calling for effective market regulation. In this paper, we consider identifying Ethereum accounts from a graph classification perspective, and propose an end-to-end graph neural network framework namedEthident, to characterize the behavior patterns of accounts and further achieve account de-anonymization. Specifically, we first construct an Account Interaction Graph (AIG) using raw Ethereum data. Then we design a hierarchical graph attention encoder namedHGATEas the backbone of our framework, which can effectively characterize the node-level account features and subgraph-level behavior patterns. For alleviating account label scarcity, we further introduce contrastive self-supervision mechanism as regularization to jointly train our framework. Comprehensive experiments on Ethereum datasets demonstrate that our framework achieves superior performance in account identification, yielding 1.13% ~ 4.93% relative improvement over previous state-of-the-art. Furthermore, detailed analyses illustrate the effectiveness ofEthidentin identifying and understanding the behavior of known participants in Ethereum (e.g. exchanges, miners, etc.), as well as that of the lawbreakers (e.g. phishing scammers, hackers, etc.), which may aid in risk assessment and market regulation.