Abstract: To explore the content characteristics of high-quality live streaming, text mining is conducted on the subtitle text. Firstly, the LDA model was used to extract themes, and the content of high-quality live streaming was divided into responsiveness, personalization, visualization, professionalism, reliability and entertainment, and the new definitions of them were given. Then, a suitable machine learning algorithm was used to predict the topic probability. Finally, the probability distribution of the content characteristics of high-quality live streaming was given by statistical analysis. The research showed that live streaming content provided utilitarian, hedonic and social value. The introduction duration of experience products was higher than that of search products, and experience products are higher than search products in terms of professionalism and entertainment, and lower in terms of visualization. In addition, the product introduction duration of the new live streaming studios was relatively shorter than that of mature ones and the new live streaming studios paid more attention to responsiveness. Some reasonable explanations were given for these findings based on previous research results. This study would help guide the work of new anchors and optimize the process for live streaming platforms.

Key words: live streaming shopping, text mining, content characteristic, LDA model