Extraction of Keywords of Novelties From Patent Claims

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 1192–1200, Osaka, Japan, December 11-17 2016. Extraction of Keywords of Novelties From Patent Claims Shoko Suzuki IBM Research - Tokyo 19-21 Nihonbashi, Hakozaki-cho, Chuo-ku, Tokyo, Japan e30126@jp.ibm.com Hiromichi Takatsuka IBM Japan 19-21 Nihonbashi, Hakozaki-cho, Chuo-ku, Tokyo, Japan e50508@jp.ibm.com Abstract There are growing needs for patent analysis using Natural Language Processing (NLP)-based approaches. Although NLP-based approaches can extract various information from patents, there are very few approaches proposed to extract those parts what inventors regard as novel or having an inventive step compared to all existing works ever. To extract such parts is difficult even for human annotators except for well-trained experts. This causes many difficulties in analyzing patents. We propose a novel approach to automatically extract such keywords that relate to novelties or inventive steps from patent claims using the structure of the claims. In addition, we also propose a new framework of evaluating our approach. The experiments show that our approach outperforms the existing keyword extraction methods significantly in many technical fields. 1 Introduction Recently there are growing needs for analyzing patents. Many companies want to analyze large amount of patents for various purposes like patent retrieval or analyzing technical trends, etc. For searching and analyzing large amount of patents, NLP-based approaches are adequate, and many approaches are developed (for example (Abbas et al., 2014)). Several keyword extraction methods are proposed in the context of patent retrieval or information extraction from patents. Most of them use traditional unsupervised approaches like BM25 (Robertson and Zaragoza, 2009) or supervised approaches like CRF (Lafferty et al., 2001). While BM25 tends to extract keywords that are characteristic to each patent, CRF is applied to extract various kinds of Named Entities such as technologies, effects, and attributes. (for example in (NTCIR, 2010),(Nishiyama et al., 2010)). However, considering the original purpose of submitting patents, patents must contain rich information not limited to the above examples. Especially, every patent must contain what the inventors think as novel or having an inventive step compared to the all existing works ever. There is no doubt that extracting such keywords is quite important and applicable to all other patent analysis like patent retrieval or analyzing technical trend, etc. To our knowledge there are very few works that explicitly try to extract those keywords that relate to the novelties or the inventive steps of each patent (we will call these keywords as keywords of novelties in this paper). In general patent retrieval task, various kind of weights are calculated for keywords/keyphrases. But these weights don’t necessarily reflect the degree of the novelties. Several approaches seem to extract novel parts of each patent implicitly, but they don’t go so far as to extract keywords of novelties. Besides, in patent retrieval, similar patents often have different surface expressions especially on the novel parts, resulting in the situation that the performance in patent retrieval task is not necessarily related to the performance of extracting the novel part. This means …

Content truncated to 500 words