| Urban management enters into the age of information and people raise issues abouturban management through the Internet. Through non-standard writing, texts of complaintformat are quite different. Thus, the staff must read verbatim to find important event fromexponential growth of texts of complaint about urban management, which is inefficient.By adopting the technique of information extraction, we can extract eventautomatically by converting the unstructured data into structured data. It not only improvesthe work efficiency, but also helps the urban management department to master theimplementation of policy and find the existing problems in the social management. In thispaper, we mainly focus on the research on complaint texts about urban management anddivide the issue into five parts: complaint text classification, recognition of Chineselocation entity, extraction of domain-specific terms, event recognition and normalization,statistical analysis of events.（1） The study of recognition of Chinese location entity in complaint texts. Wepropose a novel method to identify Chinese location entity based on thedivide-and-conquer strategy. Firstly, we use CRF role labeling to identify the basic placename. Secondly, by using semi-automatic way, we build indicator lexicon. Finally, wepropose attachment connection algorithm to connect the basic place name with indicator,then we achieve the identification of location entity. In brief, our method decomposeslocation entity into basic place name and indicator, which is different from traditionalmethods. Results of the experiments show that the proposed method has an outstandingeffect and the F-value gets to84.79%.（2） The study of location normalization. We first propose the research of Chineselocation entity normalization based on Baidu API. Using the Baidu open Place API tosearch defect location entity, constructing location area feature vectors based on resultsretrieved, and on its basis, we use the standardized rules to add the area where defectlocation located in and achieve location normalization. Experimental results show that theproposed method has an outstanding effect and the F-value gets to97%.（3） The study of extraction of domain-specific terms in complaint texts. Wepropose a variance-based method to extract candidate domain-specific terms in terms ofbetween domains, within a certain domain and unbalance of corpus. Then, we fill the meaningless word of candidate terms by using the combinational degree between twowords based on original text. Finally, we can obtain meaningful terms. Experiments on thecorpus of complaint about urban management show that our approach is effective to extractdomain-specific terms. We evaluate the performance of our approach manually andcompare the results against TFIDF, the Accuracylenientand Accuracystrictare13percent and7percent higher than TFIDF.（4） The study of event objects normalization. Complaint ontology about urbanmanagement is introduced to normalize words. We transfer words normalization tomapping between concepts and instances by calculating the similarity between the wordsand the keyword is mapped to the concept that similarity is high. Experimental resultsshow that our method is not only to ensure a high accuracy rate, but also solve the datasparseness in the mapping process.（5） Statistical analysis of events in complaint texts. On the basis of above studies,we analyze the data from three dimensions, namely, location, time, event objects. From thetime dimension, we find that the number of complaints increased with the growth of time;From the location dimension, we find that the number of complaints in the city is morethan that in rural areas, so it is necessary to strength the propaganda of rural areas;Comprehensive analysis of the three dimensions, we find the time of complaint mostlyfrom March to October. So authorities can give different emphasis depending on theseason. |
