Point-Wise Approach for Yandex Personalized Web Search Challenge

The paper describes a solution for the Yandex Personalized Web Search Challenge. The goal of the challenge is to rerank top ten web search query results to bring most personally relevant results on the top, thereby improving the search quality. The paper focuses on feature engineering for learning to rank in web search, including a novel pair-wise feature, shortand long-term personal navigation features. The paper demonstrates that point-wise logistic regression can achieve the stat-of-the-art performance in terms of normalized discounted cumulative gain with capability to scale up.