摘要：

Abstract:

【Purpose/significance】Clustering network public opinion events not only makes the public opinion information more hierarchi⁃

cal and organized, but also assists in follow-up research such as personalized public opinion event recommendation.【Method/process】Com⁃

bining Network Representation Learning (NRL) with K-means, the purpose of public opinion events clustering is achieved through four stag⁃

es: public opinion collection, event co-occurrence frequency analysis, event dimensionality reduction mapping and cluster analysis. After ob⁃

taining the public opinion event, the event co-occurrence matrix is obtained according to the co-occurrence relationship between events, and

the NRL correlation algorithm is used to obtain the low-dimensional vector representation of the public opinion event. Then K-means is

used to determine the number of groups, the initial cluster is divided, updating the category center based on the mean represented by the

low-dimensional vector of events in the category; and finally, the iteration to clustering is completed.【Result/conclusion】Using the 220 lyr⁃

ic events classified by the ant square public opinion monitoring software, it is found that K-means clustering integrated into NRL can

achieve better clustering effect.【Innovation/limitation】Based on the mining of public opinion events, we innovatively propose a k-means

clustering method that integrates network representation learning to obtain clear public opinion events. However, the amount of data avail⁃

able for personal research is limited, and it is difficult to achieve the optimal clustering effect. Internet information platforms have massive us⁃

er data, which can achieve better clustering results for personalized recommendations and other follow-up research.