Post-Learning Optimization of Tree Ensembles

Learning to Rank (LtR) is the machine learning method of choice for producing highly effective ranking functions. However, efficiency and effectiveness are two competing forces and trading off effectiveness for meeting efficiency constraints typical of production systems is one of the most urgent issues. This extended abstract shortly summarizes the work in [4] proposing CLEaVER, a new framework for optimizing LtR models based on ensembles of regression trees. We summarize the results of a comprehensive evaluation showing that CLEaVER is able to prune up to 80% of the trees and provides an efficiency speed-up up to 2.6x without affecting the effectiveness of the model. Modern search engines are expected to return highly relevant results in a fractions of seconds to satisfy efficiency constraints. Learning-to-Rank (LtR) [1] methodologies are nowadays pervasively used as effective solutions to ranking problems. However, efficiency and effectiveness are intertwined concepts than often counteract each other. In this extended abstract we shortly summarize the work in [4] where we introduce CLEaVER, a framework developed on top of QuickRank [5], for the optimization of LtR models based on ensembles of regression trees after the learning phase has completed. Since document scoring cost by using a tree ensemble model is linear in its size, CLEaVER first removes a subset of the trees, and then fine-tunes the weights of the remaining ones according to a given quality measure. Results of a comprehensive evaluation using QuickScorer [2, 3], a state-of-the-art algorithm for efficient scoring, show that CLEaVER is able to improve the efficiency of a given ranking ensemble up to a 2.6x speed-up factor without affecting the effectiveness of the model.