The paper introduces a Cognitive Personalized Search (CoPS) model that integrates large language models (LLMs) with an efficient memory mechanism to enhance user modeling in personalized search. Traditional search engines often provide identical results for all users, ignoring individual preferences. Personalized search aims to address this by re-ranking search results based on user preferences. The CoPS model leverages LLMs, known for their ability to perform complex tasks without task-specific fine-tuning, to improve user modeling in zero-shot scenarios. The CoPS model incorporates three main components: a cognitive memory mechanism, an LLM, and a ranker. The cognitive memory mechanism includes sensory memory, working memory, and long-term memory, inspired by the human brain's memory system. Sensory memory identifies re-finding behaviors, working memory integrates historical information, and long-term memory stores user preferences. The LLM processes these inputs to model user intent, while the ranker determines the relevance of documents based on user preferences. The paper discusses the challenges of data sparsity and the complexity of user histories, which are addressed by the external memory mechanism. This mechanism allows the LLM to efficiently access and process relevant segments of user history without processing the entire history. The experimental results show that CoPS outperforms existing zero-shot personalized search methods, demonstrating its effectiveness in personalizing search results without relying on extensive training data. The paper also explores the impact of different memory units and history lengths on performance, highlighting the importance of each component. Additionally, it analyzes the role of LLMs in query rewriting, user profile retrieval, and user modeling, confirming their critical functions in the personalized search pipeline. The efficiency of the model is evaluated through fine-tuning and inference, showing that CoPS achieves comparable performance to fine-tuned models while being more efficient. Finally, the paper discusses future improvements, including personalized GPT tuning and further optimization of the memory mechanism. The study concludes by emphasizing the potential of CoPS in enhancing personalized search results and the need for further research in privacy protection and security issues.The paper introduces a Cognitive Personalized Search (CoPS) model that integrates large language models (LLMs) with an efficient memory mechanism to enhance user modeling in personalized search. Traditional search engines often provide identical results for all users, ignoring individual preferences. Personalized search aims to address this by re-ranking search results based on user preferences. The CoPS model leverages LLMs, known for their ability to perform complex tasks without task-specific fine-tuning, to improve user modeling in zero-shot scenarios. The CoPS model incorporates three main components: a cognitive memory mechanism, an LLM, and a ranker. The cognitive memory mechanism includes sensory memory, working memory, and long-term memory, inspired by the human brain's memory system. Sensory memory identifies re-finding behaviors, working memory integrates historical information, and long-term memory stores user preferences. The LLM processes these inputs to model user intent, while the ranker determines the relevance of documents based on user preferences. The paper discusses the challenges of data sparsity and the complexity of user histories, which are addressed by the external memory mechanism. This mechanism allows the LLM to efficiently access and process relevant segments of user history without processing the entire history. The experimental results show that CoPS outperforms existing zero-shot personalized search methods, demonstrating its effectiveness in personalizing search results without relying on extensive training data. The paper also explores the impact of different memory units and history lengths on performance, highlighting the importance of each component. Additionally, it analyzes the role of LLMs in query rewriting, user profile retrieval, and user modeling, confirming their critical functions in the personalized search pipeline. The efficiency of the model is evaluated through fine-tuning and inference, showing that CoPS achieves comparable performance to fine-tuned models while being more efficient. Finally, the paper discusses future improvements, including personalized GPT tuning and further optimization of the memory mechanism. The study concludes by emphasizing the potential of CoPS in enhancing personalized search results and the need for further research in privacy protection and security issues.