Effective Power Market Data Classification: method, dataset and classification framework based on BERT

In the era of big data, with the exponential growth of data, data governance has become a key focus for enterprises and industries. As a major aspect of data governance, data classification can facilitate data management and utilization. However, the field of the power market lacks mature data classification methods and results. In addition, existing research on data governance and classification often requires manual labor. This paper proposes a hierarchical data classification model considering the business scenarios and characteristics of the power market. Secondly, a power market data inventory method is introduced which systematically inventory data through three steps: metadata inspection, data traceability, and business traceability. By converting the inventory results into features, a power market data classification dataset is constructed. Finally, this paper presents a power market data classification framework based on improved BERT and Bi-LSTM. The framework effectively extracts semantic features and keyword features from metadata. By integrating these features, it enhances the accuracy and robustness of classification. Experimental results demonstrate that the model significantly outperforms other baseline models in classification performance, with overall precision, accuracy, and F1-score reaching 0.94, 0.92, and 0.93, respectively.