Patent search is a tricky business. If an invention is truly new, then the vocabulary used in its patent application is probably also still emerging. The terms used in the patent may be very different than common terms used to describe the invention or product a few years later.

In addition, some patent applications are written with deliberately obscure language, to maximize the difficulty of searching for them. This is done, for example, to create submarine patents that remain hidden for years until they can be brought to trial for maximum financial return.

This paper addresses the problem of how to search for patent applications, in light of these difficulties.

Query expansion, for example, via pseudo-relevance feedback (PRF), is one existing technique used to expand keywords for searches of this kind. This paper presents alternatives, which the authors claim are superior to PRF. The key idea is to utilize Wikipedia as an external resource, in conjunction with WordNet, by expanding phrases based on context provided by Wikipedia page categories and titles.

This paper lists a large selection of related work and experimental results. The latter appear promising and the authors plan further experiments to improve them.