Generating Spoofing Tweets considering Points of Interest of Target User

Personal information of legitimate users shared on social networking services (SNS) can be used for identity spoofing. The simplest approach is to clone the profile information of the target user. Recent deep learning techniques have enabled us to even automatically generate spoofing messages by imitating the past messages of the target user; however, such message generators can only be trained for target users who have posted sufficient number of messages to train the generator. Further, since the legitimate users actually exist in the real world, their messages are often related to the situations in the real world. Such relations to the real world have not been considered in generating the spoofing messages, which can be the cues for detecting the identity spoofing. This paper further examines the possibility of identity spoofing even for target users who have posted only a limited number of messages based on the assumptions that messages about semantically related points of interest (PoIs) in the real world can be similar regardless of users. Our proposed method firstly collects messages about various PoIs posted by arbitrary users and estimates the semantic topic of each PoI based on the content of its messages. A topic-based message generator trained on the collected messages can be commonly used to generate spoofing messages about PoIs in the real world according to the interest of each target user.