Simulating group-level user behavior enables scalable counterfactual evaluation of merchant strategies without costly online experiments. However, building a trustworthy simulator faces two structural challenges. First, information incompleteness causes reasoning-based simulators to over-rationalize when unobserved factors such as offline context and implicit habits are missing. Second, mechanism duality requires capturing both interpretable preferences and implicit statistical regularities, which no single paradigm achieves alone. We propose Policy-Guided Hybrid Simulation (PGHS), a dual-process framework that mines transferable decision policies from behavioral trajectories and uses them as a shared alignment layer. This layer anchors an LLM-based reasoning branch that prevents over-rationalization and an ML-based fitting branch that absorbs implicit regularities. Group-level predictions from both branches are fused for complementary correction. We deploy PGHS on Meituan with 101 merchants and over 26,000 trajectories. PGHS achieves a group simulation error of 8.80%, improving over the best reasoning-based and fitting-based baselines by 45.8% and 40.9% respectively.

Multimodal knowledge editing (MKE) aims to efficiently rectify outdated or incorrect knowledge in multimodal large language models (MLLMs) while preserving reliability, generality, and locality. Recently, in-context editing (ICE) has emerged as a prevalent paradigm for MKE, focusing on inference-time context manipulation. While ICE mitigates the side effects of intrinsic interventions on MLLMs, it still suffers from an out-of-control limitation. Specifically, previous methods rely excessively on the implicit contextualization of MLLMs and regard the MKE process as a matter of chance. Based on this observation, we further identify the corresponding challenges as localizing the responsible key units and defining the triggering conditions within MLLMs. To address these challenges, we pursue a controllable ICE approach and refer to the multimodal neurons. Consequently, we propose a neuron-level controllable ICE framework for MKE, namely NeuCon-ICE. It consists of (1) a multimodal contextual neuron identification module that aims to determine where to edit by identifying the tightly coupled multimodal contextual neurons, and (2) a context-aware neuron editing module that aims to determine how to edit by selectively injecting context-aware updates into the identified neurons. Experiments on three representative MLLMs (BLIP-2, MiniGPT-4, and LLaVA 1.5) on ComprehendEdit and E-VQA demonstrate that NeuCon-ICE consistently achieves state-of-the-art overall performance, delivering an overall gain of at least 10.79% across baselines and datasets. The code is available at https://github.com/jc4357/NeuCon-ICE.

Large language models (LLMs) excel when user goals are clearly specified, yet real-world queries are often vague and evolving, forcing LLMs to guess and leading to misaligned responses. Existing approaches attempt to clarify user intents through iterative questioning. While effective in alleviating disambiguation, this paradigm tends to merely provide standard and normal responses, which fails to meet the growing demand for diverse personalized expression of users. Based on the observation, we first identify its problem as the ignore of users' mental states, in which the complicated mental elements, unclear functional rules, and evolving mental states pose obstacles to approach the problem. Therefore, in this paper, we introduce the theory of the mentalizing network in the human brain and propose a neurobiologically inspired framework, i.e., NeurPIU, that endows LLMs with human-like mentalizing capabilities for personalized intent understanding. NeurPIU constructs an intent neural network that organizes users' long-term mental states into a three-layer graph; retrieves query-relevant states via spreading activation mechanism with temporal decay; and injects an encoded cognitive prefix into a frozen LLM through a lightweight LoRA-based cognitive model to guide response generation. The network is incrementally updated after each interaction to track evolving user cognition. Extensive experiments on four benchmarks show that NeurPIU consistently improves long-term dialogue quality, especially its plug-and-play feature, and generalizes to conversational recommendation and mental health counseling. A user study with physiological measurements further indicates that NeurPIU reduces interaction time by 53.95% while improving user experience ratings by 22.38%. All data and code are released.

Large Language Models are rapidly emerging as web-native interfaces to social platforms. On the social web, users frequently have ambiguous and dynamic goals, making complex intent understanding—rather than single-turn execution—the cornerstone of effective human-LLM collaboration. Existing approaches attempt to clarify user intents through sequential or parallel questioning, yet they fall short of addressing the core challenge: modeling the logical dependencies among clarification questions. Inspired by the Cognitive Load Theory, we propose Prism, a novel framework for complex intent understanding that enables logically coherent and efficient intent clarification. Prism comprises four tailored modules: a complex intent decomposition module, which decomposes user intents into smaller, well-structured elements and identifies logical dependencies among them; a logical clarification generation module, which organizes clarification questions based on these dependencies to ensure coherent, low-friction interactions; an intent-aware reward module, which evaluates the quality of clarification trajectories via an intent-aware reward function and leverages Monte Carlo Sample to simulate user-LLM interactions for large-scale, high-quality training data generation; and a self-evolved intent tuning module, which iteratively refines the LLM's logical clarification capability through data-driven feedback and optimization. Prism consistently outperforms existing approaches across clarification interactions, intent execution, and cognitive load benchmarks. It achieves state-of-the-art logical consistency, reduces logical conflicts to 11.5%, increases user satisfaction by 14.4%, and decreases task completion time by 34.8%. All data and code are released.

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.

Retrieval-augmented generation (RAG) has demonstrated strong potential in enhancing large language models (LLMs) for complex, real-world question answering. However, existing RAG frameworks remain inadequate for temporal scenarios, primarily due to their inability to jointly model temporal constraints in both retrieval and reasoning. On the retrieval side, traditional approaches focus on semantic similarity, often returning outdated or temporally misaligned evidence. On the generation side, these systems frequently produce factually incorrect or hallucinated answers when confronted with incomplete or temporally inconsistent information. Motivated by the observed limitations, we propose ChronoReflect+, a temporal logic-aware RAG framework that incorporates hybrid temporal-aware retrieval and progressive multi-step reflection. Our method iteratively refines both retrieval and reasoning, identifying and bridging information gaps as context accumulates. Extensive experiments demonstrate that ChronoReflect+ significantly outperforms state-of-the-art RAG baselines-improving end-to-end accuracy by 15.2%-particularly on questions involving implicit time expressions and multi-hop reasoning.

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.

The enhance of accuracy in reasoning results of LLMs arouses the community's interests, wherein pioneering studies investigate post-hoc strategies to rectify potential mistakes. Despite extensive efforts, they are all stuck in a state of resource competition demand ing significant time and computing expenses. The cause of the situation lies in the failure of identifying the fundamental feature of the solutions in this line, coined as the self-denial of LLMs. In other words, LLMs should confidently determine the potential existence of mistakes and carefully execute the targeted correction. As the whole procedure conducts within LLMs, supporting and persuasive references are hard to acquire, while the absence of specific steps towards refining hidden mistakes persists even when errors are acknowledged. In response to the challenges, we present PSSD, which refers to and implements the human psyche structure such that three distinct and interconnected roles contribute to human reasoning. Specifically, PSSD leverages the recent multi-agent paradigm, and is further enhanced with three innovatively conceived roles: (1) the intuition-based id role that provides initial attempts based on benign LLMs; (2) the rule-driven superego role that summarizes rules to regulate the above attempts, and returns specific key points as guidance; and (3) the script-centric ego role that absorbs all procedural information to generate executable script for the final answer prediction. Extensive experiments demonstrate that the proposed design not only better enhance reasoning capabilities, but also seamlessly integrate with current models, leading to superior performance.

Large language models (LLMs) have shown impressive success in various applications. However, they encounter issues in accurately understanding user intentions, thereby impeding the successful accomplishment of tasks. The pioneering study tackles intention understanding through iteratively interacting with users to enhance response quality; however, it fails to identify the notorious challenges associated with the task, where efficiency and accuracy are paramount for ensuring optimal user experience. To address these challenges, we introduce a new interactive table based intention understanding (ITIU) framework, which refers to and implements non-linear thinking in psychology such that details of intention are parallelly generated. Specifically, in the table interacting design phase, ITIU first brainstorms a more concrete intention table relevant to user instructions and subsequently incorporates a rule-based supervision mechanism to enhance the accuracy of its content. In the specialized model training phase, we obtain the procedural records generated by ITIU to develop a specialized upstream interactive intention understanding model. The specialized model replaces internal steps within the original interaction design for further efficiency improvement. Comprehensive experimental results demonstrate that ITIU significantly outperforms existing intention understanding methods, particularly in terms of interaction efficiency and intention understanding accuracy. Furthermore, whether integrated into the open-source LLaMA or powerful LLMs like GPT-4 and Claude-3, ITIU shows significant performance improvements. All the data and codes are released.

Large language models (LLMs) possess powerful contextual comprehension capabilities and have demonstrated remarkable success in conversational tasks. However, existing works that apply LLMs to conversational text-to-SQL task have the problem of repetitive mistakes, which results in the failure to bring out the performance of LLMs. In this paper, we propose a novel approach that provides guidance through learning from mistakes. Specifically, the guidance offered by our approach includes tailored suggestions, corrective feedback, and personalized strategies aimed at improving learning outcomes. Furthermore, we employ chain-of-thought (CoT) to utilize guidance that is not suitable directly as prompts. Our method rigorously analyzes actual errors and strategizes on how to utilize the derived guidance effectively. Experimental results demonstrate that our approach improves the state-of-the-art (SOTA) performance metrics, increasing QEX performance from 66.3% to 70.9% (an absolute improvement of 4.6%) and IEX performance from 37.4% to 45.1% (an absolute improvement of 7.7%) on the CoSQL dataset.

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.

Retrieving event instances from texts is pivotal to various natural language processing applications (e.g., automatic question answering and dialogue systems), and the first task to perform is event detection. There are two related sub-tasks therein-trigger identification and type classification, and the former is considered to play a dominant role. Nevertheless, it is notoriously challenging to predict event triggers right. To handle the task, existing work has made tremendous progress by incorporating manual features, data augmentation and neural networks, etc. Due to the scarcity of data and insufficient representation of trigger words, however, they still fail to precisely determine the spans of triggers (coined as trigger span detection problem). To address the challenge, we propose to learn discriminative neural representations (DNR) from texts. Specifically, our DNR model tackles the trigger span detection problem by exploiting two novel techniques: 1) a contrastive learning strategy, which enlarges the discrepancy between representations of words inside and outside triggers; and 2) a Mixspan strategy, which better trains the model to differentiate words nearby triggers' span boundaries. Extensive experiments on benchmarks-ACE2005 and TAC2015-demonstrate the superiority of our DNR model, leading to state-of-the-art performance.