Cross Platform Text Mining Based on Public Emergency—Using Word2vec Model and K-means Algorithm

摘要

In recent years, with the increasing popularity of the Internet, the number of Internet users has reached 854 million in China, and social platforms have diversified and developed rapidly. Internet users can express their opinions on social hot spots through various ways. This paper takes word2vec model and K-means algorithm as the core, crawls three kinds of texts from two platforms: Bilibili.com and Zhihu.com, and collects 28816 texts with more than 1.35 million words. Through cleaning and noise reduction, Jieba package carries out word segmentation, word frequency calculation for preliminary topic analysis. Furthermore, word2vec model is constructed, and K-means algorithm is combined with human-computer cooperation to achieve more accurate text clustering. Results the user expression differences of different platforms were compared, which can be used for public opinion evolution analysis. Accurate user feedback can promote the output of high-quality content and increase the user stickiness of the platform. The improved k-means algorithm also improves the credibility of text clustering.