[Purpose/Significance] Unsupervised method of patent entity extraction is capable of addressing the issue of previous methods that are highly dependent on labeled resources, thus promoting the widespread of artificial intelligence technology in the intellectual property field and improving the ability of patent information service.[Method/Process] By combining the inherent technology classification codes of patent documents with topic modeling, this study proposed a new method that utilizes patent classification code to guide the topic allocation process in patent text, thus extracting entities without annotation dataset. [Result/Conclusion] To demonstrate the advantages of our method, the empirical analysis was conducted using a patent dataset from the field of thin-film magnetic heads in hard disk drives, along with the IPC technology classification system. The experimental results show that there is a significant difference in the performance of entity extraction for different levels of technology classifications. Moreover, the entity extraction performance based on the fifth-level IPC technology classification code is far superior to the conventional Subject-Action-Object (SAO) method.