This work addresses correlation bias and causal effect confounding in advertising recommendation systems and presents a causal learning–based recommendation framework. We first examine the limitations of conventional recommendation algorithms in complex advertising environments, where confounding variables and exposure bias often prevent models from capturing users’ true preferences. To tackle these issues, we design a unified embedding architecture that jointly represents user, advertisement, and contextual features, and incorporates a structural causal graph to explicitly model dependencies among variables. During model training, causal consistency regularization and inverse propensity weighting are integrated to mitigate the impact of biased exposure mechanisms and non-uniform sampling. A joint optimization objective is further formulated to couple click-through rate prediction with causal consistency estimation, enabling robust causal effect learning without sacrificing predictive accuracy. Extensive experiments on large-scale advertising datasets demonstrate that the proposed approach consistently outperforms several representative baselines in terms of Precision@10, Recall@10, NDCG@10, and MAP, while exhibiting strong robustness under multi-dimensional sensitivity analysis. Overall, this study highlights the practical value of causal modeling and consistency-aware learning in advertising recommendation and offers a computationally grounded approach for improving both interpretability and fairness in recommendation systems.

This study focuses on the problem of autonomous learning for intelligent agents in open-world environments and proposes an agent algorithm framework oriented toward self-exploration and knowledge accumulation. The framework couples hierarchical perception modeling, dynamic memory structures, and knowledge evolution mechanisms to achieve an adaptive closed loop from environmental perception to decision optimization. First, a perception encoding and state representation module is designed to extract multi-source environmental features and form dynamic semantic representations. Then, an intrinsic motivation generation mechanism is introduced, enabling the agent to maintain continuous exploration even without external rewards, thus promoting active discovery and accumulation of knowledge. Meanwhile, a jointly optimized policy network and knowledge updating module is constructed, allowing the agent to continuously integrate new experiences and refine old knowledge during long-term interactions, forming a stable and scalable knowledge structure. Experimental results show that the model achieves superior performance in uncertainty suppression, policy consistency maintenance, and behavioral deviation control, demonstrating its effectiveness and robustness in open-world tasks. This research enriches the theoretical foundation of autonomous learning and provides a feasible technical pathway for building general intelligent systems with self-driven and continuously evolving capabilities.

Large Language Model (LLM)-based multi-agent systems have demonstrated remarkable capabilities across di- verse applications, yet they face critical security challenges in- cluding backdoor attacks, prompt injection, and privacy leakage. Existing defense mechanisms typically address single threat vec- tors, lacking a unified governance architecture for comprehensive security. We propose SecureGov-Agent, a governance-centric multi-agent framework that introduces a dedicated Governance Agent responsible for monitoring inter-agent communications, auditing tool invocations, and enforcing security policies. Our framework incorporates a multi-perspective risk scoring mech- anism that evaluates content risk, privacy risk, and behavioral anomalies to dynamically assess each agent’s trustworthiness. We further enhance robustness through adversarial training on syn- thesized attack scenarios. Extensive experiments across medical consultation, financial advisory, and document processing scenar- ios demonstrate that SecureGov-Agent achieves a balanced trade- off between security, privacy, and efficiency: reducing attack success rates by 73.2% compared to unprotected systems and privacy leakage rates by 81.4%, while maintaining 89.7% task completion rate with only 15.3% latency overhead. Notably, our framework excels in privacy protection (6.8% leakage rate) and maintains practical efficiency, offering a comprehensive solution for privacy-sensitive multi-agent deployments. Our framework provides a reproducible benchmark for multi-agent security research and offers practical deployment guidelines for privacy- sensitive applications.

Large language model (LLM)-based multi-agent systems have demonstrated remarkable capabilities in collaborative task solving. Although the mechanisms that facilitate seamless cooperation, such as shared contexts, role assignments, and iterative message passing, present significant risks of unintentional information disclosure. We present MIN-Trust, a trust orchestration framework that enforces Minimum Necessary Information (MNI) constraints, an operationalization of the data minimization principle for inter-agent communication—while maintaining task effectiveness. Our approach introduces an MNI-Gate that automatically classifies and filters information into essential, summarized, or pointer-referenced subsets before transmission. Additionally, we propose a Trust-Gated Channel (TGC) that counterintuitively increases verification requirements rather than relaxing information access as inter-agent trust elevates. Through experiments on four collaborative tasks using public benchmarks, we demonstrate that MIN-Trust reduces sensitive information exposure by 67.8% compared to baseline multi-agent frameworks while maintaining 93.3% of task success rates. Our evidence traceability mechanism achieves 84.2% claim-to-source attribution, significantly outperforming conventional approaches. These results suggest that privacy-preserving multi-agent collaboration is achievable under synthetic benchmark conditions with moderate performance trade-offs.

To address the limited controllability, unstable output consistency, and weakly constrained decision processes of large language models in text classification tasks, this work proposes a controllable prompt-driven text classification method that establishes an end-to-end unified modeling framework from instruction alignment to constrained decoding. Text classification is reformulated as an instruction-conditioned generative discriminative problem. Input texts and task instructions are jointly encoded to form a unified internal representation that integrates textual semantics with classification constraints. On this basis, a category semantic alignment mechanism is introduced to ensure that the model explicitly follows category boundaries and decision criteria defined by the instructions, thereby reducing classification inconsistency caused by prompt variation or implicit bias. To further improve output reliability, a structured constrained decoding strategy is designed to restrict the generation space to a predefined set of valid categories, preventing redundant text or invalid outputs from interfering with classification results. Comparative analysis under unified data and evaluation settings demonstrates that the proposed method achieves more consistent advantages in classification accuracy, discriminative stability, and overall separability. These findings indicate that deeply integrating instruction understanding and output control into the decision process of large language models effectively transforms their generative capacity into stable, interpretable, and controllable text classification capability, providing a systematic solution for building reliable intelligent text analysis systems.

This study addresses the challenge of long-term dependency modeling in agent behavior planning for long-horizon tasks and proposes a memory-driven agent planning framework. The method introduces hierarchical memory encoding and dynamic memory retrieval structures, enabling the agent to selectively retain and effectively utilize historical information across multiple time scales, thereby maintaining policy stability and goal consistency in complex dynamic environments. The core idea is to construct an interaction mechanism between short-term and long-term memory, where attention-guided retrieval integrates historical experience with current perception to support continuous planning and decision optimization in long-term tasks. The proposed framework consists of four key modules: perception input, memory encoding, state updating, and behavior generation, forming an end-to-end task-driven learning process. Experimental evaluations based on success rate, average planning steps, memory consistency score, and policy stability demonstrate that the proposed algorithm achieves superior performance in long-term task scenarios, effectively reducing planning redundancy and improving strategy coherence and task efficiency. The results confirm that the memory-driven mechanism provides a novel theoretical foundation and algorithmic framework for developing long-term task agents, establishing a solid basis for adaptive decision-making and continuous planning in complex environments.