In recommendation tasks, most existing models that learn users’ preferences from user–item interactions ignore the relationships between items. Additionally, ensuring that the crossed features capture both global graph structures and local context is non-trivial, requiring innovative techniques for multi-scale representation learning. To overcome these difficulties, we develop a novel neural network, CoGraph, which uses a graph to build the relations between items. The item co-occurrence pattern assumes that certain items consistently appear in pairs in users’ viewing or consumption logs. First, to learn relationships between items, a graph whose distance is measured by Normalised Point-Wise Mutual Information (NPMI) is applied to link items for the co-occurrence pattern. Then, to learn as many useful features as possible for higher recommendation quality, a Convolutional Neural Network (CNN) and the Transformer model are used to parallelly learn local and global feature interactions. Finally, a series of comprehensive experiments were conducted on several public data sets to show the performance of our model. It provides valuable insights into the capability of our model in recommendation tasks and offers a viable pathway for the public data operation.

Large Language Models (LLMs), now a foundation in advancing natural language processing, power applications such as text generation, machine translation, and conversational systems. Despite their transformative potential, these models inherently rely on massive amounts of training data, often collected from diverse and uncurated sources, which exposes them to serious data security risks. Harmful or malicious data can compromise model behavior, leading to toxic outputs or hallucinations, while also creating vulnerabilities to data-driven attacks such as prompt injection and data poisoning. As LLMs continue to be integrated into critical real-world systems, understanding and addressing these data-centric security risks is imperative to safeguard user trust and system reliability. This survey offers a comprehensive overview of the main data security risks facing LLMs and reviews current defense strategies, including adversarial training, data cleaning, output guardrails, Reinforcement Learning from Human Feedback (RLHF), data augmentation, and Retrieval-Augmented Generation (RAG)/agent defenses. Additionally, we categorize and analyze relevant datasets used for assessing robustness and security across different domains, providing guidance for future research. Finally, we highlight key research directions that focus on data provenance and traceability, verifiable machine forgetting, secure model updates, standardized evaluation framework, explainability-driven security analysis, and effective governance frameworks, aiming to promote the safe and responsible development of LLM technology. This work seeks to inform researchers, practitioners, and policymakers, driving progress toward data security in LLMs.