Our student Hahn Thi Hong joined the TextGraphs-16 Shared Task on Natural Language Premise Selection. Given a mathematical statement written in natural language as the input, we try to create a system to predict the relevant premises that could support an end-user finding proof for that mathematical statement.
By an empirical evaluation of several contextual representations relying on Transformer-based language models as well as evaluation of the performance of different similarity scores, we achieved the #rank 1 above all the competition in the task. The paper will be published soon under the 16th Workshop on Graph-Based Natural Language Processing (TextGraphs 2022).
