The Cooperative Patent Classification (CPC) provides a standardised and reliable framework for organising patent information. As a structured taxonomy, CPC offers not only consistency for retrieval but also a basis for conducting technology trend analysis. This study develops a method for detecting and visualising technological shifts by leveraging the structural features of CPC. Compound words are extracted from patent claims and evaluated through two unsupervised measures: reconstruction errors calculated with AutoEncoder and anomaly scores estimated by an Isolation Forest. Based on these complementary measures, a four-quadrant model is developed to classify patents according to the degree and nature of technological change. Furthermore, a time-series analysis is conducted for CPC categories positioned in the quadrant with both high reconstruction errors and high anomaly scores, which represent areas of pronounced technological change. Many of the top-ranking words in 2020 show continuous growth since 2016, confirming that these terms are expanding within heterogeneous CPC areas identified through the model. The proposed method provides a systematic framework for identifying emerging technologies and heterogeneous technical elements, offering a robust basis for understanding technology evolution. It also demonstrates practical potential for supporting corporate R&D strategy, technology management, and forecasting of future technological priorities.
野田正貴 et al. (Tue,) studied this question.