Improving Events Classification with Latent Space Clustering-Based Similarities

The research on event intelligent analysis based on big data refers to the intelligent classification of monitoring events through the analysis of monitoring event alarm information in the operation and maintenance platform, to automatically recommend monitoring event processing solutions according to the event knowledge base. However, there are currently few methods to classify monitoring events. To solve this problem, our method relies on the BERT model and the Jieba word segmentation tool to perform keyword extraction, keyword word vector transformation and event representation vector generation for event information in training data. We then pre-classify the training data using the clustering algorithm and similarity to obtain information about each cluster. We establish the relationship between clusters and event classifications based on the pre-classification results and the classification labels of the training data. Finally, we process and analyze the new monitoring events that appear in the operation and maintenance platform, and effectively classify the new events according to the model training results. In addition, our method can periodically train the model to optimize the classification performance based on dynamically added data from the monitoring event database. We perform experiments on the real-life datasets and the results validate the effectiveness of our proposed method.