AdaGP-Rank: Applying boosting technique to genetic programming for learning to rank

One crucial task of learning to rank in the field of information retrieval (IR) is to determine an ordering of documents according to their degree of relevance to the user given query. In this paper, a learning method is proposed named AdaGP-Rank by applying boosting techniques to genetic programming. This approach uses genetic programming to evolve ranking functions while a process inspired from AdaBoost technique helps the evolved ranking functions concentrate on the ranking of those documents associating those `hard´ queries. Based on the confidence coefficients, the ranking functions obtained at each boosting round are then combined into a final strong ranker. Experiments conform that AdaGP-Rank has general better performance than several state-of-the-art ranking algorithms on the benchmark data sets.