Automatic Discovery of Prior Art : Big Data to the Rescue of the Patent System

In this research, we offer a fresh approach as to determining prior art. We do this by using Big Data methods. More specifically, we apply a model which constructs the semantic space of patents, in which all published patents and patent applications are arranged according to semantic similarities between each other. Our model provides a clear indication of how closely patents stand in relation to existing technologies, which we refer to as Near Inventions (“NI”). Our model exposes a certain level of deficiency when it comes to the disclosure, by patent applicants, of NIs. One conclusion which we draw from this approach is that there is no consistency among applicants when it comes to citing NIs. Another conclusion is that the more “densely populated” the semantic neighborhood of an invention is, the more rigorous the examination needs to be regarding its patentability