Abstract
The implementation of big data-based analysis, conducting forecasts, and increasing service capabilities in smart government affairs can effectively enhance the efficiency of urban emergency management. The government hotline is an effective citizen relationship management (CRM) tool that generates a large amount of appeal information every day. Identifying and predicting public opinion hotspots of citizen complaints in real time and accurately identifying periodic and mass incidents present major challenges to city managers. In this study, we propose a pattern-based identification and early warning system for public opinion. This system uses an improved mining algorithm for frequent patterns to accurately identify topics and their corresponding case information and adopts a word-weight method to assign weights to frequent patterns, such that more important information would have a higher weight. Cosine similarity is used to calculate a similarity matrix for frequent patterns in the text content, thereby accurately distinguishing between repeated events and events of major interest. Moreover, a hash table-based retrieval and improved text rank are proposed to extract text summaries. Finally, we define sudden issues of major interest and develop an identification and early warning system for public opinion that is easy to operate, with effective user interaction and data visualization interfaces. A real case study is implemented to experimentally identify public opinion accurately and perform early. The average accuracy rate of data mining reached 87.95% in the first half of the operational evaluation of the system. Besides, when compared with the analysis of the conventional SQL statements, the retrieval efficiency is improved by 6 times and supports multi-keyword retrieval. Consequently, the enhanced text rank summary extraction algorithm improves p@10 accuracy by 6%.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call