Supervisor: Ilaria Tiddi (i.tiddi@vu.nl)
Despite recent advancements in NLP, large language models cannot still solve Math World Problems (MWPs) with high accuracy. To tackle this challenge, this project will look into (1) creating a Knowledge Graph of MWP, by designing the ontology and linking the existing dataset to Semantic Web resources, and (2) create a graph-to-text dataset for GNN training. Thereafter, either one can either work on new methods for solving these problems or on developing a benchmark for graph2text methods (by using the ability of LLMs to solve the generated MWP from the graph as accuracy metric).