Researchers have developed a method for training knowledge bases using supervised structure learning. In this approach, an agent is trained to answer questions based on a document store, which it then edits and updates. The result is a more accurate and efficient knowledge base that outperforms unsupervised methods by a significant margin. The study also introduces a new metric called key-coverage gradient, which measures how well the training set generalizes to unseen questions. --- Why it matters: This matters because it provides a more effective way to build and maintain large-scale knowledge bases, which are essential for many AI applications such as question answering and text generation. Source: https://arxiv.org/abs/2608.21829