Query Enhancement for Patent Prior-Art-Search Based on Keyterm Dependency Relations and Semantic Tags

Abstract

Prior art search is one of the most common forms of patent search, whose goal is to find patent documents that constitute prior art for a given patent being examined. Current patent search systems are mostly keyword-based, and due to the unique characteristics of patents and their usage, such as embedded structure and the length of patent documents, there are rooms for further improvements. In this paper, we propose a new query formulation method by using keyword dependency relations and semantic tags, which have not been used for prior art search. The key idea of this paper is to make use of patent structure, linguistic clues and use word relations to identify important terms. Moreover, to formulate better queries we attempt to identify what technology area a patent belongs to and what problems/solutions it addresses. Based on our experiments where IPC codes are used for relevance judgments, we show that keyword dependency relation approach achieved 13~18% improvement in MAP over the traditional tf-idf based term weighting method when a single field is used for query formulation. Furthermore, we obtain 42~46% improvement in MAP when additional terms are used through pattern-based semantic tagging.