Patent Classification with Intelligent Keyword Extraction

Nowadays, companies invest to promote innovative ideas to have the edge over its competitor. These upcoming ideas are comprehensively defined in patent documents which are readily available in the public domain. So, there is a need to analyze patent documents to achieve a strong market position, get high returns on investment, and identify new business segments. One popular method for analyzing patent documents is manually classifying each technical or scientific document into several predefined technical categories by field experts. However, this manual classification approach is expensive in terms of time, cost and it is error-prone. Also, there is a requirement for extended efforts for handling frequent data updates. In contrast, cheaper and faster operations are enabled by Artificial Intelligence techniques and can relieve the human resources burden. In this paper, we suggested an intelligent keyword extraction technique to help business professionals easily identify technologies and labels of sub-technologies involved in the patent document. In this research, we considered 35,477 patent documents from the commercial patent database. We implemented an intelligent keyword extraction technique to obtain meaningful keyword sets associated with technical information from patent documents. Later on, we trained Google's BERT (i.e., Bidirectional Encoder Representations from Transformers) keyword extraction model on textual input (title, abstract, and claims) and keyword sets from patent documents for predicting patent technology and sub-technology labels. Afterward, the performance of the proposed method is compared with K-means clustering+ TF- IDF and LDA-based topic modeling. The experimental outcomes illustrate that our proposed algorithm offers a reasonable means to classify patent documents by extracting dominant keywords from patent texts. With the proposed approach, we achieved 97.18% accuracy for patent technology identification.