Research on power marketing data mining and clustering techniques based on Bert and k-meas

Clustering analysis is an important branch of data mining, which is applied to the power industry to improve the market competitiveness of power enterprises. This paper proposes a machine recognition algorithm KBert (Bert+K-Means) for specific type clustering of power marketing texts. The algorithm first converts the power marketing text into a high-dimensional text matrix; Secondly, iteratively optimizes key weight parameters in the Chinese Bert model to obtain a global semantic vector. Finally, in order to solve the limitations of the traditional BERT model, we introduced the K-Means algorithm to improve. The results show that the proposed KBert model overcomes the problems of long distance of power marketing text and uneven classification of sample types, and the performance index F1 value is better than the traditional BERT and Attention+Bilstm models, which realize the fast clustering recognition of multiple power marketing information with high accuracy.