Publications
カンファレンス (国際) Can a Machine Reading Comprehension Model Improve Ad-hoc Document Retrieval?
Kota Usuha (Tsukuba univ.), Makoto P. Kato (Tsukuba univ.), Sumio Fujita
The 24th International Conference on Asia-Pacific Digital Libraries (ICADL 2022)
2022.12.7
We propose a method to solve ad-hoc document retrieval tasks using a reading comprehension model. To solve the ad-hoc retrieval task, the proposed method generates a question for the given query, and a reading comprehension model is employed to determine whether the target document contains a corresponding answer to the generated question, thereby estimating the relevance of the document. Experimental results show that a simple application did not improve the performance in ad-hoc retrieval tasks. Through extensive analysis of the experimental results, however, we found that the proposed method was effective for improving the performance when it was applied to queries containing proper nouns.
Paper :
Can a Machine Reading Comprehension Model Improve Ad-hoc Document Retrieval?
（外部サイト）