An Overview of Learning to Rank for Information Retrieval

This paper presents an overview of learning to rank. It includes three parts: related concepts including the definitions of ranking and learning to rank; a summary of pointwise models, pairwise models, and listwise models; estimation measures such as Normalized Discount Cumulative Gain and Mean Average Precision, respectively. Considering the deficiency that current learning to rank models lack of continual learning ability, we present a new continual learning idea that combines multi-agent autonomy learning mechanism with molecular immune mechanism for ranking.