To address the low efficiency of feature mining and limited prediction accuracy in enterprise service user intent prediction, a research proposes an enterprise service user intention prediction model that integrates heuristic variants inspired by Kmeans++and Stacking ensemble learning. The model improves traditional K-means++ clustering through adaptive weighted grid information entropy optimization, solving the problems of slow convergence and uneven weight distribution in large-scale data. It also builds a weighted ensemble learner using base classifiers such as random forest to enhance intent prediction performance after multidimensional feature fusion. The experimental results show that the optimized Fast K-means++clustering algorithm achieved a contour coefficient of 0.92, a Calinski Harabasz index of 2500, and a Davies Bouldin index of 0.12 on dense point datasets, with significantly better clustering quality than the comparative algorithms. In the testing of the FK Stacking prediction model in real e-commerce scenarios, the accuracy, recall, and F1 score all exceeded 0.97, and the error rate remained stable below 2.1% in medium and long-term time series predictions; After iterative optimization, the model’s memory usage was reduced by 50% and response time was shortened by 82.5%. The results show that the proposed model offers lightweight and high-accuracy advantages in enterprise service user data analysis and intent prediction. It can help enterprises optimize resource allocation and improve service response speed.
Han et al. (Mon,) studied this question.