A Multi-Channel Retriever for Effective Academic Question Answering

Fuxin Jiang, Quanying Lu

Keywords: Multiple-channel Retriever, Ensemble

Abstract:

In this era of booming technology and rapidly updated informa-
tion, it has become a top priority to provide researchers and the
general public with high-quality cutting-edge academic knowledge
in multiple fields. Accurate academic paper retrieval can help re-
searchers quickly capture the frontiers of their fields and accelerate
research progress. For this purpose, we propose a multi-channel
retriever that includes a Naïve Embedding-based Retriever and a
Graph Embedding-based Retriever that considers the citation rela-
tions between papers. We then use Reciprocal Rank Fusion (RRF)
to ensemble the results from the multiple retrievals. Our approach
achieved fifth-place position in the KDD Cup 2024 Task 2 competi-
tion.

Submission Number: 25