Searching for Prior-Art Patents

The goal of this project is to study automatic methods for finding related patents based on a
combination of many features extracted from the target patent and, thus, allow to address the
task of finding prior art for a patent (application). In this setting, the patent document is used as
the query for finding prior art. Retrieval is accomplished by building feature sets including
languages models, n-grams, etc. of both, the query patent and the patents in a large patent
corpus. The approach employs ranking algorithms such as RankNet to discover highly associated
resources.