Patent Value Score Prediction Based on BERT-XGBoost-Stacking with Late Fusion

Abstract

Predicting the value scores of high-value patents is essential for evaluating technological innovation, managing intellectual property, and promoting industrial development. However, existing methods still face challenges in effectively fusing multimodal data and achieving high prediction accuracy. To address this issue, we propose a late fusion model based on BERT-XGBoost-Stacking to improve the accuracy and robustness of patent value assessment. This approach leverages BERT to extract deep semantic features from patent texts, employs XGBoost to model structured numerical features, and optimizes the fusion strategy through the Stacking framework. Experimental results on patents in China's new energy vehicle sector demonstrate that the proposed method outperforms single-modal models in terms of Mean Squared Error (MSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE), indicating higher prediction accuracy and stability. This study not only enriches quantitative methods for patent valuation but also provides new insights for the application of artificial intelligence in patent analysis, thereby supporting technological innovation and high-quality industrial development.