Patents are the most effective carriers of technical information and have become key assets in the innovation competition between enterprises and countries. Universities are a major focus area for the National Intellectual Property Administration's crackdown on "abnormal" patents. While most university patents in China tend to be shelved after the grant of patent rights, and the investment in university scientific research resources has not produced practical application value, resulting in serious waste. Both the crackdown on abnormal patents and the transformation and utilization of patents are directly related to patent quality. Thus, it is urgent to evaluate the quality of university patents. However, it is difficult to assess the quality of patents manually through human means amidst millions of patent applications. Patent quality changes with market changes and technological development, so the dynamic nature of patent quality assessment can reflect the latest changes in the market and technology in a timely manner, which can improve the accuracy of patent assessment. Therefore, it is particularly important to seek a dynamic assessment method that is both accurate and generalizable to judge the quality of patents.
In order to achieve accurate assessment of the quality of announced patents, this study constructs a dynamic patent quality assessment model based on deep learning. The model takes into account both the static and dynamic characteristics of patents, including multiple indicators in legal, market, and technological aspects, making the assessment results more comprehensive and objective. In terms of static assessment, indicators include the remaining validity period of the patent right, the number of claims, and the number of examples, which reflect the basic attributes and legal value of the patent. By constructing a static assessment quality interval, the quality status of the patent is clarified at the static level. In terms of dynamic assessment, the market transaction situation and technological development trends of similar patents are considered, including indicators such as average monthly transaction price, average monthly transfer price, transaction volume, and technical feature indicators. These indicators help capture the performance and technological influence of patents in the market, thereby better reflecting the dynamic quality of patents. By integrating static and dynamic assessments, an open deep learning assessment model is constructed to provide the corresponding quality assessment interval according to the classification numbers. This not only improves assessment efficiency but also provides strong data support for the transformation and application of patents.
Through empirical analysis, this study uses the announcement patent data from the National Intellectual Property Administration to select 1 000 samples for the extraction of static assessment quality intervals and dynamic assessment quality intervals. By establishing distribution interval charts for static and dynamic assessments and using specific screening methods, high and low screening intervals are determined, and corresponding assessment quality intervals are calculated. On this basis, the results of static and dynamic assessments are integrated to construct a deep learning assessment model for accurately assessing the quality of announced patents. The model considers multi-dimensional factors such as the technical level of patents, degree of innovation, and market demand, and can output three levels of assessment grade signals: normal, low, and high. It also uses a redirection mechanism for further analysis of abnormal or borderline assessment results. In addition, the model's training used a BP neural network, including an input layer, a hidden layer, and an output layer, where the hidden layer was designed with two intermediate layers and three neuron units to enhance the model's processing capability and precision. With continuously updated patent data, the model can learn and adjust itself to improve the accuracy and reliability of the assessment.
The dynamic patent quality assessment model based on deep learning proposed in this study is not only innovative in theory but also has high practical application value. It provides a new perspective and method for patent assessment, and helps to promote the effective transformation and rational use of patents. Future research can further optimize the model structure and explore more valuable assessment indicators to improve the model's assessment precision and practicality.