Enhancing Search Efficiency through LLM-Based User Memory Systems for Query Matching and Intent Modeling

Traditional search engines primarily rely on keyword matching and ranking algorithms, which often fail to capture users’ implicit intents and contextual needs. This paper presents an LLM-based search framework that integrates user memory and behavioral modeling to enable proactive, context-aware retrieval. By continuously analyzing user interaction patterns such as past queries, click behavior, and temporal preferences the system builds dynamic user profiles that guide the generation of adaptive query embeddings. This approach allows the model to infer what users intend to search, rather than what they type, resulting in faster response times and significantly higher relevance in returned results. Experimental evaluations demonstrate that the proposed LLM-memory framework reduces query latency by 21.8% and improves top-1 precision by 15.6% compared to traditional retrieval systems. The study highlights the potential of user memory augmented LLMs to reshape search paradigms, bridging the gap between explicit queries and latent human intentions.