Optimizing learning to rank models with neural network-based feature selection techniques

Learning-to-rank (LTR) is a very important task in information retrieval. However, in today’s era of exponentially growing data volumes available over the internet, acquiring highquality data for feature selection poses a significant challenge in LTR. The emerging method of employing neural network-based feature selection in learning to rank utilizes machine learning methodologies to identify and choose relevant features, to enhanced efficiency and improve the ranking. This strategy overcome the constraints of conventional LTR methods and enhances the ranking performance across diverse applications. This research presents a new iteration of a sequential neural network model through the utilization of existing models, along with an analysis of its attributes. A distinctive approach to feature selection utilizing neural networks is outlined in this research for learning to rank, employing LambdaMART, a gradient boosting algorithm, to enhance the effectiveness and efficiency of ranking protocols on extensive datasets. The proposed method performance is better than the few existing state-of-arts algorithms. The model performance found as 0.711 NDCG, 0.61 P@K,0.611 on MSLR Web 30k and proposed model useful in many real time ranking applications.