Application of the LDA-LSTM-Seq-BERT Model in Identifying Prime Core Patents

The evolution of natural language processing technologies has facilitated the classification and recognition of patent texts. While a plethora of methodologies exists to achieve this aim, many still possess inherent limitations. This paper introduces a novel model, LDA-LSTM-Seq-BERT, amalgamating LDA, LSTM, and BERT, devised to enhance the quality of patent text classification and bolster recognition accuracy. Initially, an extensive array of patent text data was amassed from multiple patent databases, followed by meticulous preprocessing and label extraction. Subsequently, LDA was employed to perform thematic clustering on the patent classification label set, and further training was executed using LSTM. In Culmination, we procured a high-caliber technical patent recognition set, known as “bottleneck” technology. This investigation furnishes a groundbreaking and efficacious approach to patent classification and recognition.