Research On Event And Figure Profile For Enterprise Public Opinion Monitoring

Compared with other types of public opinions,enterprise public opinions require higher timeliness of problem solving and strong decision and analysis capacity.On the one hand,enterprise public opinions usually erupt centrally in a short time,leaving limited time for decision-makers,especially for listed companies,which often suffer from economic losses.On the other hand,the spread source is uncontrollable.Compared with the government public opinions with strict public opinion control mechanism,the sources of enterprise public opinions are very wide and without any restrictions.Various forms of communication lead to tired or slow responding of enterprises.Therefore,the research on monitoring enterprises public opinions has important social significance and commercial values.The existing research on public opinion monitoring focuses on a single direction,either on thematic and emotional monitoring,or on the trend of public opinion evolution.There are few studies attempting to integrate multiple methods.However,ignoring any kind of characteristics will interfere with the accuracy of public opinions judgment and decision-making.The emergence of entity profile technology provides a new way to solve this problem.On the one hand,through the study of various public opinion events and modeling,and the integration of various research theories and methods in the field of public opinion,rich public opinion labels are extracted to assist public opinion management departments in multi-dimensional combination analysis,decision-making and governance.On the other hand,the promotion of entity profile in the field of public opinion is conducive to the in-depth study of the behavior preferences of specific users or groups.However,there were no deep and systematic studies in this area which lead to lack of practical significance.Based on this,this paper puts forward the research methods of event and figure profile for enterprise public opinion monitoring,which consists of two types of profile:event profile and figure profile.Event profile depicts each public opinion event as an object,and integrates topic/sentiment model and information dissemination theory to convert the specific public opinion events into multiple public opinion feature labels,and constructs an event profile system for enterprise public opinion monitoring.Figure profile is an insight into the characteristics of participants behind public opinion events from the user’s perspective.We adopt a variety of semantic mining algorithms and logistic regression models to depict the high-risk group behind the events based on historical behavior of users.The ultimate goal is to locate and identify high risk groups and reduce the incidence rates.Finally,the experiment on the dataset of microblog was carried out to validate the event and figure Profile model we proposed which had promising results.Meanwhile,the KS value and AUC value of the model anticipating high-risk groups are 0.7472 and 0.9412,which demonstrate that the model has good discrimination and can accurately identify high-risk groups.This work was supported by the Major Project of the Ministry of Education of China(Grant No.17JZD034).