Abstract: ［Purpose/Significance］Webcast platform is a new kind of social media platform that has been developing rapidly in recent years and increasingly favored by users. Analyzing users’ information interaction behavior on webcast platform is helpful to better understand users’ behavior characteristics and their topic preferences. ［Design/Methodology］With the use of natural language processing technology and entity discovering technology, this study builds a characteristic model for the topic map of users’ information interaction behavior on live webcast platform, and conducts a topic mining. Combined with the topic on Tik Tok, this article analyzes the characteristics of users’ information interaction from three aspects, which are single user’s information interaction map, inter-group users’ interaction map and users’ information perception semantic map.［Findings/Conclusion］Users’ information interaction behavior on live webcast platform can be divided into single user’s  information interaction behavior and group users’ information interaction behavior. Mass media and self-media are the most influential webcast platforms for netizens. The characteristics of users’ information interaction behavior on live webcast platforms include information interaction motivation, isolation and bias. ［Originality/Value］This study builds a characteristic model of users’ information interaction behavior on live webcast platforms, which could help reveal the main characteristics of users’ information interaction behavior on webcast platforms better. Meanwhile, it proposes a new research perspective of semantic mining for subsequent research on topic discovery.It also proposes comments and advices for providing users with better communication and public opinion guidance on webcast platforms.

Keywords: Live webcast, Information interaction behavior, Topic map, Characteristic model, Semantic mining