Evaluating the quality of public geocoding services for crime analysis in China

doi:10.21203/rs.3.rs-1989875/v1

Download PDF

Method Article

Evaluating the quality of public geocoding services for crime analysis in China

https://doi.org/10.21203/rs.3.rs-1989875/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background

Crime researchers often use publicly available geocoding services to obtain crime locations and conduct subsequent analysis. The quality of these geocoding platforms has not been extensively investigated, especially in the crime research field. The match rate is often employed for this purpose, but this measures only absolute performance at a specified scale and cannot reflect the performance of a platform at different accuracy levels. By expanding match rate to multiple scales, we compared the quality of publicly available geocoding services in China.

Methods

In this study, we develop a set of evaluation metrics by clarifying the definition of the match rate and considering the multiscale characteristics of geocoding errors. To interpret the geocoding errors of addresses, the positional errors are classified into more detailed types in accordance with their topological relationships with mapping units. Using burglary addresses recorded in N city, the quality of the geocoding services provided by four mainstream online geocoding services in China are compared based on these metrics.

Results

The match rates of the four geocoding platforms are high enough to maintain the burglary distribution patterns at the subdistrict and district levels but cannot satisfy this need at the building level. Three of the four platforms can geocode enough addresses for spatial analysis at the community level, while Tencent cannot. For commercial addresses, the match rates of the four platforms for residential addresses cannot satisfy the needs of burglary mapping at the building level but can satisfy these needs at any other level. For residential addresses, Tencent is the only platform that cannot provide satisfactory results for community-level mapping. For gated community-level mapping, Baidu and Tianditu can provide marginally satisfactory results. Based on the accuracy levels provided by each platform, the applicability of the results at each level are further analyzed.

Conclusions

Although this study is limited by address type and data volume, the results suggest users to select appropriate geocoding service based on mapping unit sizes and address types. The results can also help users determine whether the geocoding result at certain accuracy level meet their needs. This research also provides guidance for address-based users beyond crime analysis.

online geocoding service

geocoding errors

quality evaluation

Geocoding is the process of converting addresses into XY coordinates via a matching algorithm between a given address and a reference address database. Geocoding techniques play an essential role in various aspects of crime research, including crime mapping, spatial modeling and crime prediction. Errors in the geocoding process may be translated into uncertain points on a map and therefore introduce uncertainty into crime analysis. Moreover, some addresses cannot be identified because of a lack of information or the presence of incorrect characters. Many related research questions have been proposed, such as “Are geocoded points reliable for crime mapping?” [1] and “How do geocoding errors affect the subsequent analysis?” [2]. These questions are commonly encountered by crime researchers and practitioners and are directly related to the uncertainty of geocoding techniques.

With the increasing demand for location-based services in China, many online companies have developed online geocoding platforms and provide geocoding services for the public. However, these geocoding services are not commonly used by crime researchers because their applicability in crime research has not been seriously examined. More importantly, many studies of geocoding quality assessment have been conducted only on specific areal units for analysis, such as census boundaries [1] and street segments [3, 4]. The conclusions of a study conducted on one areal unit may not be applicable to others. There is still a lack of examination of the applicability of geocoding services for crime mapping at multiple scales. This limits the application of online geocoding platforms in the crime research field.

This study aims to provide an evaluation framework that is applicable to multiple areal unit sizes and to evaluate the accuracy of China’s online geocoding platforms. The purposes of this paper are threefold: (1) to present a framework for evaluating geocoding quality by reviewing the current metrics in the literature, (2) to evaluate the quality of online geocoding services with respect to crime data in China, and (3) to compare the results of geocoding systems for two different address types against the accuracy levels provided by references from each official website.

Literature review

In geocoding research, the metrics of positional accuracy and match rate are often employed to evaluate the quality of geocoding results.

Metrics for the geocoding process

The positional accuracy is one of the most commonly used metrics in geocoding assessment. It can be calculated as the distance between a geocoded location and the ‘true’ location represented by the corresponding address [5]. Low accuracy often indicates long distances between the geocoded locations and true locations, whereas high accuracy indicates short distances. The mean value and variance further reflect the level of geocoding accuracy [6]. However, this metric is designed to evaluate the general performance of a geocoding platform and is deficient in reflecting the impact of geocoding results on spatial statistics. On the one hand, most errors are very small, and some are almost negligible. Only a few large errors can bias the mean value and variance and reduce the reliability of the positional accuracy in describing geocoding quality [7]. On the other hand, the positional accuracy cannot reflect whether a group of geocoded points with a certain accuracy level can meet cartographic requirements at a given scale. It often occurs that a geocoded point is located within the wrong areal unit even if it is very close to the ‘true’ location [1, 8]. From this perspective, the positional accuracy is limited at reflecting the impact of geocoding results on spatial mapping and spatial analysis.

The match rate is another metric used to assess the quality of geocoding results. It refers to the percentage of correctly ‘geocoded’ addresses among all data. However, the meaning of ‘matching points’ varies in different studies. The term ‘matched data’ refers to addresses that are automatically geocoded in the work of Zhan et al. [9] and to addresses that are manually selected from among geocoded candidates in the work of Zandbergen [10]. In contrast, the match rate has been calculated as the percentage of addresses successfully placed in the correct areal units in some studies [1, 3]. This leaves indeterminate definitions of points subjected to manual inspection and points located within incorrect polygons in a geocoding process. In addition, the match rate is often calculated at a certain level, such as at the level of census tracts or street segments. It cannot reflect the quality of geocoding results at multiple levels simultaneously.

The metrics mentioned above cannot accurately reflect the potential influence of geocoding on spatial analysis due to their mode of use in the literature. This may limit the application of these metrics. In addition, the definition of a metric may vary among different studies, which may lead to problems in the statistical analysis of geocoded points with large errors. Thus, a complete and unified framework for evaluating geocoding quality is necessary for users of geocoding techniques.

Geocoding techniques in China and their application in crime research

Chinese addresses are different from those in Western countries. A Chinese address consists of the province, city, district, and subdistrict where it is located, along with other detailed information [11–13]. The administrative divisions in an address are presented in a set spatial sequence, arranged in order by administrative level. The hierarchical relationship of the division levels in various addresses is shown in the following figure (Fig. 1). The detailed information can be separated into two types: "road & house number" or "community & building number". As such, a typical Chinese address can be expressed in the form “province & city & district & subdistrict & road & house number” or “province & city & district & subdistrict & community & building number” [14].

With recent advancements in geocoding techniques, many geocoding platforms have been developed that provide application programming interfaces (APIs) for public users. Many crime researchers employ these APIs to translate addresses into (x,y) coordinates. These crime locations are then placed on a map to facilitate spatial analysis based on areal units of different sizes. It is commonly noted, however, that the geocoding process requires manual inspection [15–20]. Furthermore, these platforms present limitations in supporting fine-grained spatial analysis due to the less than perfect geocoding quality. Geocoding quality, which represents the performance of these platforms, has not yet been seriously investigated in the Chinese context. In view of the above problems, the current research will quantitatively evaluate the quality of the coordinates geocoded by four online geocoding platforms in China.

In this study, a set of metrics will be proposed to evaluate the quality of online geocoding services available in China, including the number of lost addresses, the number of addresses geocoded in incorrect units at multiple scales, and the number of addresses geocoded in correct units. To interpret the geocoding errors of addresses, the geocoded locations are associated with the true locations represented by the addresses. These associated point pairs are then classified into more detailed types in accordance with their topological relationships with mapping units (Table 1).

Table 1. Metrics used for geocoding evaluation and error types

	Match type	Topological relationship with mapping units	Metrics used to measure the error
Geocoding error	Matched with building	A, B	Match rate
	Misplaced in another unit	C, D, E, F, G	Match rate
	Lost	None	Loss rate

Loss rate

The loss rate is calculated as the percentage of lost addresses among all input addresses. In a geocoding process, the input addresses are required to follow a certain form to ensure that they can be correctly recognized by the geocoding platform. Data loss can be caused by an unstandardized input address, an incomplete database, or a less than perfect matching algorithm.

Matched data

Matched data refer to addresses matched with an address in a database. In a geocoding process, the most similar address is selected from the reference database, and the corresponding coordinates are matched to the input data. However, the matched coordinates are often not the same as the true coordinates, meaning that some finite distance exists between the geocoded location and the true location.

For commercial addresses, geocoding errors can be separated into five categories according to their spatial relationships with the mapping units. If the true location and geocoded location are within the same building, then this error will not affect the mapping accuracy. These errors are represented by types ‘A’ and ‘B’, as shown in Figure 2. The remaining geocoding errors can be separated into three cases, which are represented by types ‘C’, ‘D’ and ‘E’, as shown in Figure 2. C represents cases in which the true and geocoded locations belong to two different road areas. Such an error is often generated due to misidentification of a road name in the geocoding process. D represents cases in which the geocoded and true locations are in two different buildings but within the same roadside area. E represents cases in which the geocoded and true locations are in two different buildings located on opposite sides of the same road. Whether the three aforementioned cases affect the mapping accuracy depends on their spatial relationships with the mapping units. If the solid line polygon in the figure is the mapping unit, none of the three error types will not affect the mapping accuracy. However, if the dashed line polygon is used for analysis, then the results of spatial aggregation will be changed.

For residential addresses, geocoding errors can be classified into four categories according to their relationships with the mapping units. If the geocoded point is within the correct community, the address is matched at the community level. If the geocoded location is located within the correct district, this address is matched at the district level. These two cases are shown as ‘A’ and ‘B’ in Figure 3. These two errors will not affect the accuracy of building-level mapping and subsequent analysis. If the addresses are geocoded in the wrong building (‘C’ and ‘D’ errors in Figure 3), the impacts on the mapping accuracy will depend on their spatial relationships with the mapping units. If the mapping unit is the solid blue polygon in the figure, then this error will not affect the mapping accuracy. However, if the mapping unit is the dashed line polygon, then this error will change the spatial statistics at this level.

Acceptance level

The minimum acceptance level for the geocoding match rate has been extensively investigated in previous research. These parameters are also used to evaluate the acceptance level in this study. Ractliffe (2004) suggested 85% as the minimum acceptable geocoding match rate. Briz-Redon, et al. (2019) observed that this threshold is mostly between 80% and 90% and suggest to raise it because it is heavily sensitive to research purposes or statistical techniques. As suggested by these research results, the minimum acceptable geocoding match rate at the community level or building level is determined to be 80~90%. If a match rate is lower than 80%, the result is considered to be unable to correctly reflect the spatial distribution pattern of the input data. If the match rate is between 80% and 90%, then the result can marginally satisfy the mapping requirement. If the match rate is higher than 90%, the result is good enough to maintain the spatial distribution pattern of the input data.

Research area and data

The built area of N city was selected as the research area. N city is located in the highly developed Yangtze River Delta region. Many online companies provide location-based services in this region, including navigation, food delivery, bike sharing, etc. A large volume of addresses has been collected and stored in a reference database for public use. The highly developed online services and well-constructed database for N city make it an ideal region for geocoding research.

We use burglary addresses as the research data, which were collected by the local public security bureau (PSB). In contrast to typical point of interest (POI) addresses, the addresses in the current list were reported by the victims and verified by law enforcement in person. This process enhanced the completeness and correctness of the data.

The dataset includes 7259 burglary addresses (Figure 4). Each address is composed of four parts: “city + district + subdistrict + community/road + building number”. These addresses are categorized into two groups. The first group includes residential addresses, represented by “community + building number”. The second group includes commercial addresses, represented by “road + building number”.

The precise coordinates represented by each address were collected by the PSB and manually inspected based on the Global Positioning System (GPS). In this geocoding process, four online geocoding platforms, namely, Baidu, Gaode, Tencent and Tianditu, were used to translate these addresses into coordinate pairs. Each coordinate pair was assigned a confidence level indicating its accuracy level. WanderGIS [21] was used to transform these Mars coordinates into WGS84 coordinates due to its good performance in this transformation [21]. The geocoding error of each address was calculated from the spatial distance between the geocoded coordinates and its precise coordinates.

General analysis

Buildings, communities and subdistricts are the most commonly used mapping units for crime analysis. At the building level, Tencent could successfully geocode far fewer addresses than the other three platforms (Fig. 5). When the addresses matched at the building and community levels are considered together, Tencent could match slightly fewer addresses than Baidu, Tianditu, and Gaode. This result is different from that in a previous study, which argued that Tencent outperforms the other platforms [13]. Two reasons may account for this discrepancy. First, the previous study used an accuracy metric calculated as the distance between the geocoded and true locations to measure the quality of the geocoding results. Second, the previous study used POI addresses as the research object. Compared with burglary records, POI addresses are relatively unstandardized because they are usually registered by public users. This may introduce considerable error into the geocoding process.

Given the minimum geocoding match rates provided by previous studies, at least 80 ~ 90% of addresses should be matched to maintain the spatial patterns of burglaries at the building level and at the (gated) community level, respectively. In this experiment, the match rates of all four platforms could not satisfy the corresponding requirement at the building level (Fig. 5). Three of the four platforms could satisfy the requirement at the community level, while Tencent could marginally satisfy this requirement. At the subdistrict level, the geocoding match rates were all sufficiently high to maintain the burglary distribution pattern.

For the burglary addresses recorded in the form “road + building number”, the match rate for Tencent at the building level was lower than those of Baidu, Tianditu, and Gaode (Fig. 6). However, the cumulative match rates of the four platforms were very close at the community, subdistrict and district levels. In summary, the geocoding match rates of the four platforms for residential addresses cannot satisfy the needs of burglary mapping at the building level but can satisfy these needs at any other level.

For the burglary addresses recorded in the form “community + building number”, the match rate of Tencent was much lower than those of the other three platforms at the building level (Fig. 7). Similarly, the Tencent match rates were generally 5 ~ 10% less than those of the other three platforms at several other levels, including the gated community, community, and subdistrict levels.

The geocoding match rates of the four platforms for residential addresses are not sufficiently high to maintain the burglary distribution pattern at lower levels but can satisfy the needs for analysis at the subdistrict and district levels. The accuracy of Baidu and Tianditu geocoding can marginally satisfy the requirement for analysis at the gated community level, while Gaode and Tencent geocoding cannot satisfy this requirement. At the community level, the accuracy of Tencent geocoding can marginally satisfy the mapping requirement, while the other three platforms can fully satisfy this requirement.

In summary, the match rates of these four publicly available geocoding platforms are high enough to maintain burglary distribution patterns at the subdistrict and district levels but cannot satisfy this need at the building level. Three of the four platforms can geocode sufficient addresses for spatial analysis at the community level, while Tencent cannot. This is due to the weak geocoding ability of Tencent for residential addresses. For commercial addresses, the four platforms can geocode sufficient addresses for spatial analysis at the community, subdistrict and district levels but cannot satisfy this need at the building level. For residential addresses, Baidu and Tianditu can geocode marginally sufficient data, while Gaode and Tencent cannot geocode sufficient data.

Comparison of error type composition between commercial addresses and residential addresses

In this experiment, the misplaced commercial addresses and residential addresses were further categorized into three classes (error types C, D, and E) and two classes (error types F and G), respectively, according to their error types.

For commercial addresses, the proportions of different error types of four geocoding platforms are roughly at the same level. D error was at a very high level (46.53%~53.21%), followed by A&B error (18.57%~28.60%) and C error (17.48%~20.77%). This means that many addresses were geocoded to the wrong building or location on the right street. This might be caused by a lacking of building addresses or weak ability at matching building numbers. Many other addresses were geocoded to a wrong street. This might be caused by a lacking of street segments of the reference database or the weak ability of matching streets. E error (5.53%~6.47%) and lost addresses account for a low proportion. This means that a few addresses cannot be geocoded or geocoded to a wrong road side.

Tencent matched far fewer addresses (18.57%) than the other platforms (25.87 ~ 28.60%) (Table 2). Consequently, Tencent generated 5% more D errors than the other platforms at several levels, meaning that Tencent geocoded more addresses as belonging to other buildings on the same road segment. Gaode and Tencent generated 3% more C errors than Baidu and Tianditu. This means that Gaode and Tencent matched more addresses with incorrect road segments. The loss rates for the four platforms were all similar, at a very low level (0.99 ~ 1.85%).

Table 2

Comparison of the geocoding results with commercial addresses at the levels of various spatial units
Platform	A&B	Topological relationship with polygons									Lost
		C			D			E
		Community	Subdistrict	District	Community	Subdistrict	District	Community	Subdistrict	District
Baidu	28.60	14.81	16.07	17.49	45.45	45.26	46.53	3.11	5.12	5.53	1.85
Gaode	25.87	16.98	19.24	20.61	45.91	45.99	46.85	3.54	5.31	5.63	1.05
Tencent	18.57	17.95	19.32	20.77	51.54	51.19	53.21	4.21	6.14	6.47	0.99
Tianditu	28.60	14.81	16.07	17.49	45.45	45.26	46.53	3.11	5.12	5.53	1.85

For residential addresses, the proportions of different error types of three geocoding platforms other than Tencent are roughly at the same level. The lost addresses account for a low proportion. This means that a few addresses cannot be geocoded or geocoded to a wrong road side. G error was at a very high level (34.06%~38.67%), followed by A&B error (18.53%~46.12%) and F error (18.32%~39.84%). This means that many addresses were geocoded to the wrong building or location. This might be caused by a lacking of building addresses or weak ability at matching building numbers. Many other addresses cannot be matched with a building. This might be caused by a lacking of building numbers of the reference database or the weak ability of matching algorithm.

Tencent matched 23% fewer addresses than the other platforms at the building level (Table 3). Consequently, Tencent generated more F errors than the other platforms. This means that Tencent geocoded many addresses in incorrect polygons, such as incorrect gated communities, communities, subdistricts, and districts.

Given the error causes, the large percentage of C error means that the Tencent platform is weaker than the other platforms at geocoding residential addresses because of a lack of reference address or poor ability of the matching algorithm.

Table 3

Comparison of the geocoding results with residential addresses at the levels of various spatial units
Platform	A&B	Topological relationship with roads								Lost
		F				G
		Gated Community	Community	Subdistrict	District	Gated Community	Community	Subdistrict	District
Baidu	46.12	12.00	16.62	17.95	18.32	23.36	31.82	32.50	34.06	1.50
Gaode	41.56	12.37	18.71	20.41	21.18	23.24	34.06	34.40	35.73	1.53
Tencent	18.63	27.10	35.39	38.05	39.84	21.06	33.41	35.28	38.67	2.86
Tianditu	46.12	12.00	16.62	17.95	18.32	23.36	31.82	32.50	34.06	1.50

Comparison of geocoding errors against accuracy levels among the four platforms

For crime research, the reliability of a platform is of great concern. Each geocoding platform returns a number for each input address to indicate its accuracy level. This value is considered uniform across China according to the documentation and validations presented in previous research. In this study, the reliability was defined as the percentage of correctly geocoded addresses among all addresses at a certain accuracy level. For example, if 100 addresses were geocoded by a platform, 80 addresses were marked with an accuracy level of ‘90’, and only 60 of these were actually located within the correct community, then the reliability of the ‘90’ level results claimed by this platform would be 60/80 = 75% for community-level mapping. Each of the four platforms tested in this study uses its own accuracy system that is different from the others. Therefore, the accuracy levels claimed by these platforms will be analyzed separately (Figs. 8–15).

For commercial addresses, the accuracy levels cluster at a certain level. The accuracy levels of Baidu’s ‘80’ level, Gaode’s ‘House’ level, Tencent’s ‘9’ level, and Tianditu ‘90’ level can almost satisfy the requirements of crime mapping at the community level, subdistrict level, and district level. However, the results at these levels cannot meet the requirements of crime mapping at the building level.

For residential addresses, there is a great difference in the platforms’ abilities to satisfy the needs of crime mapping. The results geocoded by Baidu at the ‘50’, ‘70’, and ‘80’ levels can satisfy the requirements of crime mapping at the community level and higher. The results of Gaode’s ‘House’ level, Tencent’s ‘10’ level, and Tianditu’s ‘90’ level can also meet the requirements of crime mapping at the community level and higher.

Based on an analysis of the inconsistency of the metric definitions used for geocoding error measurement in existing research, we have proposed a set of metrics that can be used to evaluate the quality of geocoding results at multiple scales.

Using burglary addresses in N city, a large city in China, this study examines the accuracy of four publicly available geocoding platforms and tests their applicability for crime analysis. This is possibly the first study of the applicability of online geocoding platforms in crime research.

Among the four platforms, Tencent underperforms the other platforms. The match rates of the four geocoding platforms are high enough to maintain burglary distribution patterns at the subdistrict and district levels but cannot satisfy this need at the building level. Three of the four platforms can geocode enough addresses for spatial analysis at the community level, while Tencent cannot. This is due to the weak geocoding ability of Tencent for residential addresses.

For residential addresses, Baidu and Tianditu can geocode marginally sufficient data, while Gaode and Tencent cannot geocode sufficient data.

By separating the addresses into commercial addresses and residential addresses, the applicability of the four geocoding platforms has been further analyzed. The results can satisfy the needs of crime mapping and analysis at the community level and higher. The results of the four geocoding platforms are the same for commercial addresses but different for residential addresses. Tencent is the only platform that cannot provide satisfactory results for community-level mapping. For gated community-level mapping, Baidu and Tianditu can provide marginally satisfactory results. Error source analysis shows that Gaode and Tencent match many commercial addresses with incorrect road segments. Tencent geocodes many residential addresses in incorrect polygons, such as incorrect gated communities, communities, subdistricts, or districts. Tencent generally underperforms the other platforms at geocoding residential addresses because of a relative lack of reference addresses or poor ability of the matching algorithm.

Based on the accuracy levels provided by each official geocoding platform, the applicability of the results at each level has been further analyzed. For commercial addresses, the accuracy levels of Baidu’s ‘80’ level, Gaode’s ‘House’ level, Tencent’s ‘9’ level, and Tianditu ‘90’ level can almost satisfy the requirements of crime mapping at the community level, subdistrict level, and district level. However, the results at these levels cannot meet the requirements of crime mapping at the building level. For residential addresses, the results geocoded by Baidu at ‘50’, ‘70’, and ‘80’ levels can satisfy the requirements of crime mapping at the community level and higher. The results at Gaode’s ‘House’ level, Tencent’s ‘10’ level, and Tianditu’s ‘90’ level can also meet the requirements of crime mapping at the community level and higher.

The results suggest users to select appropriate geocoding service based on mapping unit sizes and address types. The results can also help users determine whether the geocoding result at certain accuracy level meet their needs. The conclusions of this study are limited by address type and data volume, the results may change for different research data and research area. Further analysis of various types of addresses or at other cities will be necessary to solidify the conclusions.

API

application programming interface

PSB

public security bureau

POI

point of interest

GPS

Global Positioning System

CBD

central business district

Acknowledgements

Not applicable

Authors' contributions

ZW conceptualization, formal analysis, funding acquisition, writing, editing.

YY conceptualization, original draft, performing the experiment.

All authors read and approved the final manuscript.

Ethics approval and consent to participate

Not applicable

Consent for publication

Not applicable

Availability of data and materials

The datasets analyzed during the current study are not publicly available due to confidential issues.

Competing interests

The authors declare that they have no competing interests.

Funding

This research was supported by two grants from the Humanities and Social Sciences of Ministry of Education Planning Fund, No.22YJCZH; Fundamental Research Funds for the Central Universities B220201021.

Authors' information

¹School of Earth Sciences and Engineering, Hohai University, Nanjing, Jiangsu, China.

Ratcliffe JH. Geocoding crime and a first estimate of a minimum acceptable hit rate. Int J Geogr Inf Sci. 2004;18:61-72.
Haberman CP, Hatten D, Carter JG, Piza EL. The sensitivity of repeat and near repeat analysis to geocoding algorithms. J Crim Justice. 2021;73:101721.
Andresen MA, Malleson N, Steenbeek W, Townsley M, Vandeviver C. Minimum geocoding match rates: an international study of the impact of data and areal unit sizes. Int J Geogr Inf Sci. 2020;34:1306-22.
Briz-Redon A, Martinez-Ruiz F, Montes F. Reestimating a minimum acceptable geocoding hit rate for conducting a spatial analysis. Int J Geogr Inf Sci. 2020;34:1283-305.
Whitsel EA, Rose KM, Wood JL, Henley AC, Liao D, Heiss G. Accuracy and repeatability of commercial geocoding. Am J Epidemiol. 2004;160:1023-9.
Cayo MR, Talbot TO. Positional error in automated geocoding of residential addresses. Int J Health Geogr. 2003;2:1-12.
Adepeju M, Rosser G, Cheng T. Novel evaluation metrics for sparse spatio-temporal point process hotspot predictions-a crime case study. Int J Geogr Inf Sci. 2016;30:2133-54.
Ratcliffe JH. On the accuracy of TIGER-type geocoded address data in relation to cadastral and census areal units. Int J Geogr Inf Sci. 2001;15:473-85.
Zhan FB, Brender JD, Lima ID, Suarez L, Langlois PH. Match rate and positional accuracy of two geocoding methods for epidemiologic research. Ann Epidemiol. 2006;16:842-9.
Zandbergen PA. A comparison of address point, parcel and street geocoding techniques. Comput Environ Urban Syst. 2008;32:214-32.
Institute of Surveying and Mapping Standardization, State Bureau of Surveying and Mapping, Nanjing City Property Rights Supervisory Office Residential and Real Estate Companies of Architectural Department. Spefcification for house surveying. Beijing: China: China Standards Press; 2000.
Sun C, Zhou S, Yang L. Chinese geo-coding based on classification database of geographical names. J Comput Appl. 2010;30:1953-5.
Tian Q, Gong Y, Kang M, Meng S, Du Q. A comparative evaluation of online geocoding services in China. Geomat Inf Sci Wuhan Univ. 2016;41:1351-8.
Peng M, Li Z, Liu H, Meng C, Li Y. Weighted geocoding method based on Chinese word segmentation and its application to spatial positioning of COVID-19 epidemic prevention and control. Geomat Inf Sci Wuhan Univ. 2020;45:808-15.
Wang Z, Liu X. Analysis of burglary hot spots and near-repeat victimization in a large Chinese city. ISPRS Int J Geo Inf. 2017;6:148.
Wang Z, Liu X, Lu J. Multiscale geographic analysis of burglary. Acta Geogr Sin. 2017;72:329-40.
Xiao L, Liu L, Song G, Zhou S, Long D, Feng J. Impacts of community environment on residential burglary based on rational choice theory. Geogr Res. 2017;36:2479-91.
Yue H, Zhu X, Ye X, Guo W. The local colocation patterns of crime and land-use features in Wuhan, China. ISPRS Int J Geo Inf. 2017;6:307.
Song G, Liu L, Bernasco W, Xiao L, Zhou S, Liao W. Testing indicators of risk populations for theft from the person across space and time: the significance of mobility and outdoor activity. Ann Am Assoc Geogr. 2018;108:1370-88.
Song G, Liu L, Bernasco W, Zhou S, Xiao L, Long D. Theft from the person in urban China: assessing the diurnal effects of opportunity and social ecology. Habitat Int. 2018;78:13-20.
WanderGIS. Coordtransform py[EB/OL]. 2015. https//githubcom/wandergis/coordTransform_py. Accessed 15 Aug 2022.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Evaluating the quality of public geocoding services for crime analysis in China

Status:

Version 1

Abstract

Background

Methods

Results

Conclusions

Figures

Background

Metrics for the geocoding process

Geocoding techniques in China and their application in crime research

Methods

Results

General analysis

Comparison of error type composition between commercial addresses and residential addresses

Comparison of geocoding errors against accuracy levels among the four platforms

Discussion And Conclusions

Abbreviations

Declarations

References

Additional Declarations

Status:

Version 1