Insider threat detection (ITD) is notoriously difficult: malicious actions are rare, context-dependent, and deliberately hidden within massive volumes of legitimate user behavior. Existing ITD methods rely on single- or fused-view models, which lack extensibility and therefore fail to leverage the supervisory signals from newly introduced complementary views. While ensembling is a natural next step, its direct application to ITD confronts three core obstacles: scalability bottlenecks from independently trained sub - models, semantic misalignment across heterogeneous feature spaces, and view imbalance, where strong views overshadow weaker yet informative ones. In this work, we propose Insight-LLM, the first extensible multi-view fusion framework tailored for ITD. Insight-LLM encodes each view with frozen pre-trained backbones and aligns heterogeneous representations into a unified semantic space via a lightweight ViewAdapter, enabling coherent cross-view reasoning without incurring additional training overhead. A context-adaptive fusion module dynamically re-weights views to emphasize subtle yet semantically consistent threat signals, and the fused representation is integrated with task prompts for lightweight LLM fine-tuning. Experiments on CERT datasets show that Insight-LLM improves F1 by up to 4.8% and reduces false positives by 61%, while decreasing training time per newly added view by up to 83.2% compared with the simple Ensemble method.

Dynamic link prediction aims to predict the future links in dynamic graphs. Existing generative dynamic link prediction studies utilize the global degree distribution for mitigating the over-estimation problem, which can model the time-invariant features while neglecting the time-varying features, resulting in capturing inaccurate evolution patterns. However, such time related features are intrinsically coupled, which makes simultaneously and independently modeling both features infeasible. Motivated by these issues, we propose a Time-wise causal debiasing framework (Tide) for generative dynamic link prediction, which does not resort to any extra trainable modules. Instead, to obtain the time-invariant features, we first utilize a time-invariant deconfounded learning mechanism for decoupling the prediction score with the degree distribution. To leverage the time-varying features, we intervene in the model during the inference stage by a predicted future degree distribution, aiming to make the accurate predictions for dynamic graphs. Experiments conducted on four public datasets under both inductive and transductive settings present that our Tide enhanced models can outperform their corresponding vanilla versions by up to 21.42% and 27.73% in terms of NDCG and Jaccard, respectively.

Log parsing extracts structured information from unstructured logs and serves as a fundamental pre-processing step for various log-based analytics and monitoring tasks. Recent advances have leveraged Large Language Models (LLMs) to handle log format complexities and enhance parsing performance. However, these methods heavily rely on labeled data, which is often scarce in rapidly evolving industrial systems, limiting their applicability in real-world scenarios. Moreover, the sheer volume of logs results in slow parsing and high computational costs, further hindering the deployment of LLM-based log parsing systems. To address these issues, we propose Parse-LLM, an unsupervised end-to-end log parsing framework based on LLMs Specifically, we first developed a Log Decomposer Agent that leverages Chain-of-Thought (CoT) reasoning and callable tools, enabling the LLM to autonomously separate log headers from content. Next, we introduce the Hybrid Log Partition module, which segments logs by balancing commonalities and differences. Finally, we developed a novel Variation-aware Log Parsing module that allows the LLM to harness additional supervisory signals through comparative analysis of similar logs. Comprehensive experiments conducted on large-scale public datasets show that Parse-LLM outperforms state-of-the-art log parsers in an unsupervised setting, offering an effective and scalable solution for the practical application of unsupervised log parsing.

Dynamic graph link prediction is widely utilized in the complex web of the real world, such as social networks, citation networks, recommendation systems, etc. Recent Transformer-based link prediction methods on dynamic graphs not only fail to model the fine-grained structures such as triangles with the vanilla Transformers in the graph serialization process, but also amplify the imbalanced distribution of graphs because of their over-estimation of high-degree nodes. To tackle these issues, we propose a Topology-aware Transformer on Dynamic Graph (TopDyG) for link prediction, consisting of a topology injected Transformer (Ti-Transformer) and a mutual information learning (Mi-Learning). The Ti-Transformer explores the explicit structure of serialized graphs, capturing the topological features. The Mi-Learning mines the relationship between nodes by modeling the mutual information with a prior knowledge, alleviating the over-estimation of high-degree nodes when applying the Transformer-based models for the dynamic graph link prediction task. Extensive experiments on four public datasets containing both transductive and inductive settings present the superiority of our proposal. In particular, TopDyG presents an improvement of 43.27% and 28.75% against the state-of-the-art baselines in terms of NDCG and Jaccard, respectively. The advantages are especially obvious on the high-density graphs.

Event detection (ED) is a challenging task in the field of information extraction. Due to the monolingual text and rampant confusing triggers, traditional ED models suffer from semantic confusions in terms of polysemy and synonym, leading to severe detection mistakes. Such semantic confusions can be further exacerbated in a practical situation where scarce labeled data cannot provide sufficient semantic clues. To mitigate such bottleneck, we propose a multilingual prompt learning (MultiPLe) framework for few-shot event detection (FSED), including three components, i.e., a multilingual prompt, a hierarchical prototype and a quadruplet contrastive learning module. In detail, to ease the polysemy confusion, the multilingual prompt module develops the in-context semantics of triggers via the multilingual disambiguation and prior knowledge in pretrained language models. Then, the hierarchical prototype module is adopted to diminish the synonym confusion by connecting the captured inmost semantics of fuzzy triggers with labels at a fine granularity. Finally, we employ the quadruplet contrastive learning module to tackle the insufficient label representation and potential noise. Experiments on two public datasets show that MultiPLe outperforms the state-of-the-art baselines in weighted F1-score, presenting a maximum improvement of 13.63% for FSED.

Job‐resume matching (JRM) is the core of online recruitment services for predicting the matching degree between a job post and a resume. Most of the existing methods for JRM achieve a promising performance by simplifying this task as a matching between the free‐text attributes in the job post and the resume. However, they neglect the contributions of the semistructured multivariate attributes such as education and salary, which will result in an unsuccessful prediction. To address this issue, we propose a novel approach to comprehensively explore the Internal and EXternal InTeractions for semistructured multivariate attributes in JRM, i.e., InEXIT. In detail, we first encode the key and the value of each attribute as well as its source into the same semantic space. Next, to explore the complex relationships among the multivariate attributes, we propose to hierarchically model the internal interactions among the multivariate attributes inside the job post and the resume, as well as the external interactions between the job post and the resume. In particular, a stepwise fusion mechanism is designed to respectively integrate the key embeddings and the source embeddings into the value embeddings so as to clearly indicate the key and the source of the value. Finally, we employ an aggregation matching layer to predict the matching degree. We quantify the improvements of InEXIT against the competitive baselines on a real‐world dataset, showing a general improvement of 4.28%, 4.10%, and 3.56% over the state‐of‐the‐art baseline in terms of AUC, accuracy, and F1 score, respectively.

Event representation targets to model the event-reasoning process as a machine-readable format. Previous studies on event representation mostly concentrate on a sole modeling perspective and have not well investigated the scenario-level knowledge, which can cause information loss. To cope with this dilemma, we propose a unified fine-tuning architecture-based approach ( UniFA-S ) that integrates all levels of trainings, including the scenario-level knowledge. However, another challenge for existing models is the ever-increasing computation overheads, restricting the deployment ability on limited resources devices. Hence, in this article, we aim to compress the cumbersome model UniFA-S into a lighter and easy-to-deploy one without much performance damage. To this end, we propose a sequence-aware knowledge distillation model (SaKD) that employs a dynamic self-distillation on the decouple-compress-couple framework for compressing UniFA-S , which cannot only realize the model compression, but also retain the integrity of individual components. We also design two fitting strategies to address the less-supervised issue at the distillation stage. Comprehensive experiments on representation-and-inference ability-based tasks validate the effectiveness of SaKD. Compared to UniFA-S , SaKD realizes a more portable event representation model at the cost of only 1.0% performance drop in terms of accuracy or Spearman’s correlation, which is far less than other knowledge distillation models.

Current question answering systems are insufficient when confronting real-life scenarios, as they can hardly be aware of whether a question is answerable given its context. Hence, there is a recent pursuit of unanswerability of a question and its attribution. Attribution of unanswerability requires the system to choose an appropriate cause for an unanswerable question. As the task is sophisticated for even human beings, it is expensive to acquire labeled data, which makes it a low-data regime problem. Moreover, the causes themselves are semantically abstract and complex, and the process of attribution is heavily question- and context-dependent. Thus, a capable model has to carefully appreciate the causes, and then, judiciously contrast the question with its context, in order to cast it into the right cause. In response to the challenges, we present PTAU, which refers to and implements a high-level human reading strategy such that one reads with anticipation. In specific, PTAU leverages the recent prompt-tuning paradigm, and is further enhanced with two innovatively conceived modules: 1) a cause-oriented template module that constructs continuous templates towards certain attributing class in high dimensional vector space; and 2) a semantics-aware label module that exploits label semantics through contrastive learning to render the classes distinguishable. Extensive experiments demonstrate that the proposed design better enlightens not only the attribution model, but also current question answering models, leading to superior performance.

Conventional deep learning-based Relation Classification (RC) methods heavily rely on large-scale training dataset and fail to generalize to unseen classes when training data is scant. This work concentrates on RC tasks in few-shot scenarios in which models classify the unlabelled samples given only few labeled samples. Existing few-shot RC models consider the dataset as a series of individual instances and have not fully utilized interaction information among them. Interaction information is conducive to indicate the important areas and produce discriminating representations. So this paper proposes a novel interactive attention network (IAN) which uses inter-instance and intra-instance interactive information to classify the relations. Inter-instance interactive information is first introduced to solve the low-resource problem by capturing the semantic relevance between an instance pair. Intra-instance interactive information is then introduced to address the ambiguous relation classification issue by extracting the entity information inner an instance. Extensive numerical experimental results demonstrate the proposed method promotes the accuracy of down-stream task.

Event detection classifies unlabeled sentences into event labels, which can benefit numerous applications, including information retrieval, question answering and script learning. One of the major obstacles to event detection in reality is insufficient training data. To deal with the low-resources problem, we investigate few-shot event detection in this paper and propose TaLeM, a novel taxonomy-aware learning model, consisting of two components, i.e., the taxonomy-aware self-supervised learning framework (TaSeLF) and the taxonomy-aware prototypical networks (TaPN). Specifically, TaSeLF mines the taxonomy-aware distance relations to increases the training examples, which alleviates the generalization bottleneck brought by the insufficient data. TaPN introduces the Poincaré embeddings to represent the label taxonomy, and integrates them into a task-adaptive projection networks, which tackles problems of the class centroids distribution and the taxonomy-aware embedding distribution in the vanilla prototypical networks.

Given an occurred event, human can easily predict the next event or reason the preceding event, yet which is difficult for machine to perform such event reasoning. Event representation bridges the connection and targets to model the process of event reasoning as a machine-readable format, which then can support a wide range of applications in information retrieval, e.g., question answering and information extraction. Existing work mainly resorts to a joint training to integrate all levels of training loss in event chains by a simple loss summation, which is easily trapped into a local optimum. In addition, the scenario knowledge in event chains is not well investigated for event representation. In this paper, we propose a unified fine-tuning architecture, incorporated with scenario knowledge for event representation, i.e., UniFA-S, which mainly consists of a unified fine-tuning architecture (UniFA) and a scenario-level variational auto-encoder (S-VAE). In detail, UniFA employs a multi-step fine-tuning to integrate all levels of training and S-VAE applies a stochastic variable to implicitly represent the scenario-level knowledge. We evaluate our proposal from two aspects, i.e., the representation and inference abilities. For the representation ability, our ensemble model UniFA-S can beat state-of-the-art baselines for two similarity tasks. For the inference ability, UniFA-S can outperform the best baseline, achieving 4.1%-8.2% improvements in terms of accuracy for various inference tasks.