RankGPES: learning to rank for information retrieval using a hybrid genetic programming with evolutionary strategies

Abstract

In recent years, Learning to Rank has not only shown effectiveness and better suitability for modern Web Era needs, but also has proved that it outperforms traditional ranking in terms of accuracy and efficiency. Evolutionary approach to Learning to Rank such as RankGP [37] and RankDE [3] have shown further improvement over non-evolutionary algorithms. However when Evolutionary algorithms have been applied to a large volume of data, often they showed they required so much computational efforts that they were not worth applying to industrial applications. In this thesis, we present RankGPES: a Learning to Rank algorithm based on a hybrid approach combining Genetic Programming with Evolution Strategies. Our results not only showed that it outperformed both RankGP [37] by 20% and RankDE [3] by 6% in terms of accuracy but also it showed it required significant less amount of time to converge to a near-optimal result.

A Hybrid-Box tool for Explainable and Interactive Simulated Annealing Strategy

“…Notable algorithms that fall under the listwise paradigm include ListNET, which was introduced by 11 , RankGP as elaborated upon by 12 , as well as 13 work from the same year. Additionally, the listwise approach encompasses Coordinate Ascent as outlined by 14 , AdaRank, as proposed by 15 , and RankGPES, as described in the work of 16 .…”