기술분류를 위한 특허 이미지화 및 군집화에 관한 연구

Patents contain detailed information about the developed technology. Therefore,patent analysis is used to establish R&D strategies and identify technology trends. For this analysis, patents must be assigned to several predefined technicalcategories. Patents are given unique codes corresponding to technical categories,which have been manually classified by some experts. Technology classificationperformed by hand has the disadvantage that errors can occur and cost and timeare high. Attempts are being made to classify technologies by applying quantitativemethods such as text mining, which have recently been developed, to patents. Quantitative patent classification is cost-effective and reproducible. In this paper,we propose a method of converting patent text, which is non-image data, intoimage data and classifying technology through clustering. The proposed methodperforms image data transformation on vectorized information with one-hotencoding and distributed representation techniques in text mining, and performstechnology classification by forming clusters. As a result of collecting andexperimenting with actual patent data, it was confirmed that the technologyclassification performance was improved when TFIDF and W2V were converted toimages.