When fact-checking methods based on large language models (LLMs) use external evidence to validate claims, knowledge conflicts often arise. These conflicts typically stem from inconsistencies between the external evidence and LLMs' internal pre-existing knowledge. Such an inconsistency could lead LLMs to draw incorrect answers when validating claims, especially when they are overly confident in their internal incorrect knowledge. Previous works on LLM-based fact-checking have overlooked this issue. This paper, for the first time, proposes a framework (namely KnowFC) to navigate this issue. Our key insight is dividing and adaptively utilizing the knowledge that LLMs know and do not know, thereby avoiding conflicts while enhancing the correctness and efficiency of fact-checking. Specifically, in KnowFC, we propose an adaptive retrieval method, where we train an LLM using a reinforcement learning algorithm coupled with the Dunning-Kruger effect-inspired reward mechanism to identify its knowledge boundaries through confidence calibration, thereby realizing adaptive evidence retrieval. Besides, we propose a reliable and debiased fact verification method, where we organize and construct reasoning graphs using retrieved evidence to verify claims, followed by a causal intervention method using causal mediation analysis to mitigate internal knowledge interference. Experimental results on both FEVEROUS and AVeriTeC datasets show that our method outperforms baseline methods in terms of accuracy and F1 score, while also improving fact-checking efficiency.

Open-world intent discovery is critical for task-oriented dialogue systems, where static intent taxonomies fail to capture emerging user intentions and existing methods show limited generalization beyond predefined label spaces. To address this issue, we propose Unified Generative Intent Discovery (UGID), a unified framework that reformulates intent understanding as a conditional text generation task, enabling both in-domain (IND) intent classification and out-of-domain (OOD) intent discovery within a single architecture. UGID adopts a two-stage training strategy. First, instruction-tuned supervised fine-tuning strengthens semantic discrimination among known intents. Second, a self-play reinforcement learning mechanism simulates iterative user–system interactions to explore, refine, and validate novel intent labels. In addition, a reward design combining semantic fidelity and domain relevance guides the generation process toward coherent and meaningful intent discovery. Experiments on three benchmark datasets demonstrate that UGID consistently achieves strong performance, reaching ACC scores of 79.63%, 82.70% and 87.50% on BANKING, StackOverflow and CLINC, respectively. Compared with the strongest generative baseline, IntentGPT-4, UGID further improves ACC by 14.87 and 4.21 percentage points on BANKING and CLINC, respectively. Ablation studies further verify the effectiveness of the proposed design. Overall, UGID provides an effective and scalable framework for intent understanding in dynamic open-world dialogue scenarios.

Information spreads quickly through social media platforms, especially fake news with negative or even malicious intentions. In recent years, psychological studies have found that explicit reminders of fake news would diminish its consequence. Therefore, it is crucial to identify their authenticity at an early stage to avoid serious consequences. However, existing methods for fake news detection either utilize auxiliary information including users’ profiles and related events propagation networks or require sufficient and high-quality training data, which is not suitable for early fake news detection in real. An increasing number of social media news not only involves natural language content but also visual content such as images and videos, which give us a new view of fake news detection at an early stage by multi-modal data. In this paper, we propose a Multi-modal Prompt Learning framework (MPL) based on the multi-modal pre-trained model CLIP for early detection of fake news. A learnable prompt module is developed to adaptively and efficiently generate prompt representations to boost the semantic context. MPL can be implemented in supervised or few-shot settings. Extensive experiments show that the proposed MPL obtains substantial performance and efficiency improvement for the early-stage fake news detection task. The results demonstrate that MPL performs considerably well compared to both the state-ofthe-art supervised multi-modal models and the latest promptbased few-shot multi-modal models. Especially, the high recall of fake news and the high precision of real news that MPL achieved compared to other baselines verify that it will better approach one of the motivations that providing early notification of “maybe real” or “maybe fake” with the release of the news.

Zero-shot stance detection is challenging because it requires detecting the stance of previously unseen targets in the inference phase. The ability to learn transferable target-invariant features is critical for zero-shot stance detection. In this paper, we propose a stance detection approach that can efficiently adapt to unseen targets, the core of which is to capture target-invariant syntactic expression patterns as transferable knowledge. Specifically, we first augment the data by masking the topic words of sentences, and then feed the augmented data to an unsupervised contrastive learning module to capture transferable features. Besides, to fit a specific target, we encode the raw text as target-specific features. Finally, we adopt an attention mechanism, which combines syntactic expression patterns with target-specific features to obtain enhanced features for predicting previously unseen targets. Experiments demonstrate that our model outperforms competitive baselines on four benchmark datasets.

Effectively predicting the popularity of information is a critical and challenging task in various social applications. Traditionally, Markov Chains(MCs) and more recently Recurrent Neural Networks (RNNs) have proliferated due to their ability to capture the dynamics of information diffusion. However, a simplifying assumption made by most of these models is to regard information diffusion as ordered sequences, without considering the irregular time intervals between each user. Meanwhile, there are many non-sequential and long-term dependencies between users that are difficult to capture. The above features are difficult to be captured due to the intrinsic limitations of recurrent models. To tackle these limitations, we propose a novel model TiSA (Time Interval aware Self-Attention for popularity prediction), which employs the self-attention mechanism to capture various inter-user dependencies. In addition, to consider the irregular intervals, we modified the conventional position embedding by translating time intervals into phase shifts of sinusoidal functions. Extensive experiments conducted on real-world datasets show that our model outperforms SOTA approaches and reduces the time complexity.

Anchor links connect information entities, such as entities of movies or products, across networks from different sources, and thus information in these networks can be transferred directly via anchor links. Therefore, anchor links have great value to many cross-network applications, such as cross-network social link prediction and cross-network recommendation. In this paper, we focus on studying the recommendation problem that can provide ratings of items or services. To address the problem, we propose a Cross-network Collaborative Matrix Factorization (CCMF) recommendation framework based on broad learning setting, which can effectively integrate multi-source information and alleviate the sparse information problem in each individual network. Based on item anchor links CCMF can fuse item similarity information and item latent information across networks from different sources. And different from most of the traditional works, CCMF can make multi-source recommendation tasks collaborate together via the information transfer based on the broad learning setting. During the transfer process, a novel cross-network similarity transfer method is applied to keep the consistency of item similarities between two different networks, and a domain adaptation matrix is used to overcome the domain difference problem. We conduct experiments to compare the proposed CCMF method with both classic and state-of-the-art recommendation techniques. The experimental results illustrate that CCMF outperforms other methods in different experimental circumstances, and has great advantages on dealing with different data sparse problems.

Reblogging, also known as retweeting in Twitter parlance, is a major type of activities in many online social networks. Although there are many studies on reblogging behaviors and potential applications, whether neighbors who are well connected with each other (called “buddies” in our study) may make a difference in reblog likelihood has not been examined systematically. In this paper, we tackle the problem by conducting a systematic statistical study on a large SINA Weibo data set, which is a sample of 135, 859 users, 10, 129, 028 followers, and 2, 296, 290, 930 reblog messages in total. To the best of our knowledge, this data set has more reblog messages than any data sets reported in literature. We examine a series of hypotheses about how essential neighborhood structures may help to boost the likelihood of reblogging, including buddy neighbors versus buddyless neighbors, traffic between buddy neighbors, activeness (i.e., the total number of blog messages a user sends), and the number of buddy triangles a user participates in. Our empirical study discloses several interesting phenomena that are not reported in literature, which may imply interesting and valuable new applications.