The Method Based On CRF For Recognizing And Refining Of Placename And Address From Microblog Text About Urban Fire

The timely perception of the fires urban location is not only helpful to the public's intelligent travel,but also to government department's efficient emergency response to fire accidents.Therefore,it has become a key issue.Based on the issues of the existing urban fire monitoring methods including large waste of manpower and material resources,as well as the low efficiency,this paper proposes a CRF-based city fire microblog text placename address recognition and refined processing method uses the advantages of fast propagation of Weibo data,large data volume,low cost,hidden urban fire location,and machine learning and data fusion methods,realizes the rapid perception of city fire location.Main work and results are as follows.(1)The acquisition and processing method of urban fire data on Sina Weibo was studied,with the statistical characteristics of geographical names and addresses designed and selected to obtained the Weibo data of fire accidents in Nanchang city from January 2017 to November 2019 by the web crawler technology,carry out text normalization and word segmentation processing,and selected the statistical characteristics of geographical names and addresses of words,parts of speech,boundaries,geographical names dictionary and suffixes according to the characteristics of the urban fire Weibo text.(2)The CRF model-based city fire Weibo texts recognition of geographical name and address was studied with BIEO labeling system and word-based labeling method selected to identify the hidden geographical name addresses using the CRF model based on statistical characteristics of geographical names and addresses.(3)The precise processing method of the geographical name address recognition results was studied with method for completing geographical name addresses designed based on a hierarchical geographical noun database and fire location identification method in multiple geographical name addresses with precise processing to solve the problems including part loss of hierarchical information of geographical name addresses and repeated information identified by CRF model.(4)The CRF-based geographical name and address recognition of Weibo texts and precise processing were completed by referring to 1562 urban fire microblog text data,with the method proposed in this paper verified.The experimental results turned out the highest precision rate of geographical name and address recognition based on the CRF model up to 88.95%,the highest recall rate is 90.97%,and the highest F-score is 88.85%.On those basis,we carried out precise processing including completion of geographic names and address,fire location identification so as to obtain a more complete and standardized city fire position.Its high degree of automation provide a new method for low-cost and time-efficient urban fire location monitoring compared with the traditional rule-based method.