Red Dragon AI at TextGraphs 2020 Shared Task: LIT : LSTM-Interleaved Transformer for Multi-Hop Explanation Ranking

TL;DR

This paper introduces the LIT model, an LSTM-Interleaved Transformer that enhances multi-hop explanation ranking for science questions by incorporating cross-document interactions, achieving competitive results in the TextGraphs 2020 shared task.

Contribution

The paper presents a novel LSTM-Interleaved Transformer architecture that improves multi-hop inference by leveraging cross-document interactions and prior ranking positions.

Findings

Achieved a test-set MAP of 0.5607 on the TextGraphs 2020 leaderboard.

Would have ranked third if submitted earlier.

Code is publicly available for replication.

Abstract

Explainable question answering for science questions is a challenging task that requires multi-hop inference over a large set of fact sentences. To counter the limitations of methods that view each query-document pair in isolation, we propose the LSTM-Interleaved Transformer which incorporates cross-document interactions for improved multi-hop ranking. The LIT architecture can leverage prior ranking positions in the re-ranking setting. Our model is competitive on the current leaderboard for the TextGraphs 2020 shared task, achieving a test-set MAP of 0.5607, and would have gained third place had we submitted before the competition deadline. Our code implementation is made available at https://github.com/mdda/worldtree_corpus/tree/textgraphs_2020