Neural Re-Ranking Models in Research Article Recommendations

Abstract

Refining the rank lists to ensure user satisfaction and experience is achieved through a re-ranking procedure, which is the culmination of the multi-stage recommender system. Gradation and correlation vary with its satisfaction. With the advent of deep learning approaches, neural-based re-ranking is a prominent research topic and is increasingly finding business applications in industry. To bolster comprehensive and effective solutions for future work by contextualizing the algorithms into a broader literature scope focused on re-ranking. It is done by creating a taxonomy of existing heuristics built on neural networks. After that, the terms of their evolution and main aims, which incorporate subdominant systems like network architecture, personalization strategies, and the welfare of computational complexity, are described. Additionally, an analysis of the main benchmark models of neural re-ranking is provided. It is also accompanied by detailed metrics for quantifying the model benchmarks used in the analysis. The final chapter, which closes the review, focuses on the research in question and other directions that can be pursued. The appendix contains cited works and benchmark datasets which is used in the analysis.