With the continuous development of science and technology, the pace of product iteration is accelerating, making it a critical challenge for designers to rapidly and efficiently develop high-quality products. Patent data information covers more than 90% of the world’s technical information. Through the systematic collection, organization, and analysis of patent data, designers can better understand existing technology, avoid potential infringement risks, and gain inspiration for subsequent innovative design. The traditional extraction of patent technology and efficacy words relies on manual labor, which is a huge workload and inefficient. To address this problem, a deep learning model based on the extraction of patent entities is proposed in this paper, aiming to automate the extraction of patent technology and efficacy entities. In addition, attention is given to the annotation strategy of the model dataset, with the objective of identifying the text annotation method that best aligns with the characteristics of this field. Using patents related to scissor lifts as a case study, the method achieved an average accuracy of over 70%(F1) with a dataset of only 180 texts, demonstrating the effectiveness of the proposed model and annotation strategy.