Geocoding-Feature_Selection

The growing popularity of location-based services has resulted in the enhancement of resources for producing geographical data and the development of the amount of data kept in geographic databases. The big data stored in databases is valuable for advanced geospatial analysis in several fields, including emergency responses, crime and traffic management, disease surveillance, and more. Geocoding, a crucial preprocessing step in geospatial data analysis, involves retrieving textual descriptions of locations into geographic identifiers. Nevertheless, geocoding outcomes delivered by worldwide service providers neglect various constraints related to textual data, including misspellings, abbreviations, and non-standard names. To overcome this issue, we propose a new approach for enhancing the quality of online geocoding services through the utilization of feature selection techniques. The proposed method is based on text similarity algorithms that are utilized to match the retrieved addresses. Compared to conventional geocoding outcomes, there is potential for an improvement of approximately 10% to 25% in the address-matching procedures employed in online geocoding services. The improvement was accomplished through the utilization of two feature selection methods, specifically mutual information feature selection and minimum redundancy maximum relevance, out of a total of fourteen approaches. Furthermore, the findings indicate that it is appropriate to prioritize character-based text similarity algorithms when comparing addresses retrieved from online geocoding services.

随着基于位置服务（location-based services）的日益普及，地理数据生产相关资源得到扩充，地理数据库中存储的数据量也随之增长。数据库中存储的海量地理大数据，可应用于应急响应、犯罪与交通管理、疾病监测等多个领域的高级地理空间分析，具备重要应用价值。地理编码（Geocoding）作为地理空间数据分析的关键预处理步骤，其核心是将位置的文本描述转换为地理标识符。然而，全球各类服务提供商提供的地理编码结果，往往未能兼顾文本数据存在的各类约束问题，包括拼写错误、缩写形式以及非标准地名等。为解决这一问题，本文提出一种基于特征选择技术的新方法，用以提升在线地理编码服务的质量。所提方法依托文本相似度算法，用于匹配检索得到的地址信息。相较于传统地理编码结果，在线地理编码服务所采用的地址匹配流程，其性能有望提升约10%至25%。该性能提升通过在共计十四种特征选择方案中筛选出的两种方法实现，即互信息特征选择法（mutual information feature selection）与最小冗余最大相关性（minimum redundancy maximum relevance）。此外，研究结果表明，在对比在线地理编码服务检索得到的地址时，优先选用基于字符的文本相似度算法更为合适。