A Topic Mining Method for Multi-source Network Public Opinion Based on Improved Hierarchical Clustering

Abstract

Heterogeneous network information platform contains common topics and characteristic topics. However, there is no unified standard for dividing public opinion topics. And the existing technology cannot adapt to the characteristics of the multi-source network platform well. This paper proposes a semi-supervised topic mining method. The core of this method is the semi-supervised hierarchical clustering algorithm improved from the traditional hierarchical clustering algorithm. On the basis of this algorithm, the optimization is carried out from the perspectives of model input vectorization and high-quality topic selection. Therefore, the method proposed in this paper can be effectively applied to the topic and hierarchical structure mining of short texts on multi-source network platforms with a wide range of topics, lots of text noise and a lack of grammatical norms. It accurately extracts the common topic and characteristic topic of the platform and the hierarchy between topics. Experiments show that this method can mine the topic and its hierarchy effectively, and it is better than the traditional LDA topic model in hierarchical structure mining and fine-grained topic mining. By analyzing the text data of the multi-source network platform, the thesis can dig out the topics and the hierarchical relationship among topics, which is conducive to analysis the subsequent research on theme retrieval and theme evolution. At the same time, network platform users and managers can obtain topic distribution information in a systematic and centralized manner. It is of great significance to guide the network's public sentiment and create a good network public opinion environment.