Abstract:

This study aims to construct a complete and efficient public opinion monitoring and analysis system to reduce the outbreak probability of negative public opinions. Based on the theory of event information structure, we constructed an event profile to monitor the enterprise public opinions. We adopted a variety of semantic mining algorithms and logistic regression models to depict the features of the high-risk population regarding the event based on historical behavior of users. The experiment was performed on the microblogging corpus to validate the event and figure profiling model that we proposed and achieved promising results. The KS value and AUC value of the model predicting highrisk groups are 0.7472 and 0.9412 respectively, which demonstrate that the model has good discrimination power. The research framework proposed in this study can effectively depict the characteristics of events and people related to enterprise public opinions.

Key words: User prediction, Event profiling, Public opinion monitoring, Influence prediction, Enterprise public opinion