A Semantic-Based Short-Text Fast Clustering Method on Hotline Records in Chengdu

As one of the main channels for citizens to reflect urban management issues, the urban hotline can collect many problems and occurrences about the city, such as noise nuisance and illegal buildings. However, traditional manual statistical methods have been adopting to analyze the topics of the urban hotline records, and few automatic analysis models or software on topic mining of urban hotline records has been documented. In order to automatically analyze the massive amount of information on urban hotline, we propose a semantic-based short-text fast clustering method to cluster short texts of semantic similarity to form long texts according to the similarity of semantic-based keywords set, and the latent dirichlet allocation (LDA) model is then applied to mine the topics distribution of urban hotline records. Experiments on 87,055 urban hotline records from 2017 to 2018 in Chengdu show that our approach can achieve a significantly better performance both in accurate and topic coherence than LDA method.