A Patent Keyword Extraction Method Based on Corpus Classification

Abstract

The keyword extraction of patents is crucial for technicians to master the trends of technology. Traditional keyword extraction approaches only handle short text like title or claims, but ignore the comprehensive meaning of the description. This paper proposes a novel patent keyword extraction method based on corpus classification (PKECC), which simulates the patent understanding methods of human patent examiners. First of all, a corpus classification model based on multi-level attention mechanism adopts the Bert model and hierarchical attention mechanism to classify the sentences of patent description into four parts including technical field, technical problem, technical solution, and technical effect. Then, the proposed keyword extraction method based on the fusion of BiLSTM and CRF is incorporated to extract keywords from the four parts. The proposed PKECC simulates understanding style of patent examiner by extracting keywords from the description. Meanwhile, PKECC may reduce the complexity of extracting keywords from a long text and improve the accuracy of keyword extraction. The proposed PKECC is compared with 5 traditional or state-of-the-art models and achieves better accuracy, F1 score and recall rate; its recall rate is above 62%, its accuracy reaches over 84%, and the F1 score arrives at 69%. In addition, the experimental results shows the proposed PKECC has a better universality in keyword extraction.