Ensembling Ten Math Information Retrieval Systems: MIRMU and MSM at ARQMath 2021
| Authors | |
|---|---|
| Year of publication | 2021 |
| Type | Article in Proceedings |
| Conference | CEUR Workshop Proceedings |
| MU Faculty or unit | |
| Citation | |
| Web | |
| Keywords | math information retrieval; question answering; math representations; word embeddings; ensembling |
| Description | We report on the systems that the Math Information Retrieval group at Masaryk University (MIRMU) and the team of Faculty of Informatics students (MSM) prepared for task~1 (find answers) of the ARQMath lab at the CLEF conference. We have prototyped ten math-aware information retrieval (MIR) systems for the main question-answering task. We ensembled the results of the ten ``weak'' individual systems into committees and let them vote to provide answers to questions. We evaluated the proposed individual systems and ensembles, considering their diversity, hyperparameters, and representations used, and classified their approaches. We have shown the diversity of all systems and evaluated four voting algorithms to collect and rank the answers. Ensembling techniques consistently outperformed the base systems and showed the power of voting of diverse systems. Our prototypes help to understand the challenging problems of question-answering in the STEM domain and our novel reproducible evaluation framework sets a new direction in MIR research. Finally, we formulate ten commandments for future work in the area. |
| Related projects: |