Researchers have developed a new approach to patent prior-art retrieval called Sparse Coverage. This method maps local spans of text in patents to a set of sparse, semantic centers. These centers are selected based on their ability to cover the most technical components and functions mentioned in the patent. Experiments show that Sparse Coverage is competitive with or even outperforms dense retrieval methods for certain tasks. The approach combines local semantic evidence with sparse inverted-index search to provide an effective first-stage retrieval method. --- Why it matters: This matters because patent prior-art retrieval is a critical task for researchers and engineers working in areas like intellectual property law, innovation management, and technology transfer. Effective retrieval can help identify relevant prior art, reducing the risk of patent infringement and facilitating more informed decision-making. Source: https://arxiv.org/abs/2608.16918