Reconciliation of Mental Concepts with Graph Neural Networks

In the digital age, knowledge processes can be formalized and simplified using task management systems. As they evolve, so must the underlying schemata to retain harmony and concurrency with the real world. In this work we present a graph neural network model that can help in reconciling these data. It can do so by leveraging a novel propagation rule that does not presume reciprocal dependency but is able to represent it still. Thereby it can predict structures in the form of usage links with high accuracy and assist in the reconstruction of missing information. We evaluate this model on a new knowledge management dataset and show that it is superior to traditional embedding methods. Further, we show that it outperforms related work in an established general link prediction task.