Enhancing Documents with Multidimensional Relevance Statements in Cross-Encoder Re-ranking

By Rishabh Upadhyay et al

Published on June 19, 2023

Read the original document by opening this link in a new tab.

Table of Contents

1. Introduction
2. Related Work
3. The CErel.stat Model for Re-ranking
4. Experimental Evaluations

Summary

This paper proposes the CErel.stat model for multidimensional relevance enhancement in cross-encoder re-ranking. It introduces a novel approach to consider additional dimensions of relevance beyond topicality, focusing on credibility. The model enhances retrieved documents with relevance statements related to credibility scores, leading to improved re-ranking effectiveness. Experimental evaluations on health-related datasets show significant performance gains over baseline models, validating the effectiveness of the proposed approach.