Da-Young Lee  Hwan-Gue Cho

10.5626/JOK.2021.48.6.639

chat message cross-texting honorifics sentence completeness sentence vector

Cross-texting refers to accidentally sending a message to an unintended person. It occurs frequently when users chat with multiple counterparts at the same time. Messengers mainly provide a function of canceling sending, but it is only post solution, and users find it difficult to prevent mistakes in advance. In this paper, we proposed a cross-texting detection model by analyzing the syntactic characteristics of chat sentences. It modelizes the previous chat messages of a specific user by extracting the honorifics and completeness features from chat messages, and detects the cross-texting cases by determining whether the target sentences are in accordance with the user chat message model. This approach is significant as it solves the cross-texting detection problem only with syntactic characteristics without semantic analysis by modeling the consistency of the user"s chat attitude. The proposed model detect cross-texting cases with an accuracy of 85.5% from automatically generated data using a real messenger dialogue corpus.