Abstract
The three dimensions of Function, Structure, and Purpose are fundamental to patent classification and play a decisive role in improving the accuracy of patent information categorization. By providing targeted abstracts for each of these three dimensions, strong summarization and contextual information can be generated, thereby enhancing the effectiveness of patent text analysis. Leveraging three-dimensional features alongside contextual information significantly improves the precision of patent classification. In alignment with the characteristics of the technical domain and the IPC classification system, this paper proposes a Patent Multilevel Domain Information (PMDI) model, designed to facilitate the targeted extraction of 3D information from patents. The PMDI model effectively captures the core 3D features necessary for classification and subsequently integrates these features into a Multi-Information Processing (MIP) model. This MIP model links the extracted 3D features with the IPC classification framework, resulting in a notable improvement in patent classification accuracy. Empirical results indicate that the summary information extracted by the PMDI model enhances classification accuracy by up to 5.67% compared to conventional deep learning approaches. When integrated with the MIP model, the classification accuracy of the multi-dimensional technology domain classification method reaches an impressive 96.77%. The proposed PMDI model, MIP model, and classification method based on structured patent text features collectively contribute to a substantial improvement in classification performance, offering significant support for knowledge-driven services such as knowledge retrieval and patent management.