Behavior-Driven Semantic Re-Ranking for Personalized Web Search

Abstract

As the web continues to expand, users face growing challenges in locating relevant and personalized information. Traditional search engines, while powerful, often deliver generic results that fail to account for individual user preferences, leading to inefficient search experiences. This study proposes a personalized re-ranking system that enhances standard search engine results by incorporating user behavior and semantic analysis. Leveraging BERT embeddings and KeyBERT for user profiling, the system dynamically reorders Google Search results to better align with users' historical interests and current queries. A user study involving 14 participants demonstrated that the proposed system significantly improves search efficiency, reduces effort, and enhances result relevance. These findings contribute to the development of adaptive search interfaces that personalize web search in real time without altering the core search infrastructure.