♣ Gómez-Barrón, J.P.; Manso-Callejo, M.A.; Alcarria-Garrido, R.P.; Gómez-Pérez, R., 2015, "Volunteering assistance to online geocoding services through a distributed knowledge solution", in RICH-VGI: enRICHment of volunteered geographic information (VGI): Techniques, practices and current state of knowledge, Workshop at the 18th AGILE Conference on Geographic Information Science, June 9th, 2015, Lisboa, Portugal.

1. 1 Introduction
In recent years there exists a growing impulse in different
areas like business, marketing or public management and
services to position geospatial technologies as an efficient
way to integrate, visualize and analyse spatial data to answer
questions with a location perspective and to obtain
knowledge. Geocoding process, consist on assigning a
geographic coordinate pair to a particular place by comparing
its descriptive location elements with those in a reference
database [1, 2, 3]. Then it is needed to search in the reference
data to assign a score to each potential candidate, filter them
based on the minimum match score and deliver the best match
[3].
In the last years, main companies that offer digital mapping
services like Google, Yahoo or Here Maps and open-data
mapping platforms like OpenStreetMap, are improving their
web service technologies and APIs (Application Programming
Interface) to tackle geocoding complexity and to make it
transparent to end users. End users must analyse the quality of
the geocoded results for each service to choose the best option
to their applications [4] and data characteristics. Service
providers are responsible for maintaining the reference
matching data and for improving predefined algorithms, so
the user cannot customize the geocoder settings or rules to
manipulate the response according to their needs or specific
input data.
While geocoding services give immediate output with high
match rates, basic user knowledge and low or no cost,
sometimes their characteristics produce low quality results,
mainly when we work with ambiguous input. Most of these
online services provide a value of results quality, like the
calculation method used or obtained entity type. These values
can be helpful as a guideline to understand the output and for
data quality assessment, complementing data quality common
metrics like completeness, positional accuracy, repeatability
[3] and similarity [4].
In the scenario of unstructured named places (addresses) as
input, the variety of online geocoding services response can
be an advantage. The platform presented in this paper
proposes to combine and analyse different geocoders outputs
as options for incomplete or imprecise data. Based on
crowdsourcing geospatial data [5, 6] and Volunteered
Geographic Information [7] approaches, the platform
facilitates the online assistance of users to analyse quality and
geographic precision in geocoding results. Identify and save
the selected best candidate or geocode manually the address,
relying on the cognitive abilities and local knowledge of the
collaborators. An address or location represents the entry of a
geocoding task and is completed by distributed users,
therefore a comparative evaluation can be made using the
platform database.
The rest of the article is organized as follows. The next
section describes the platform components and development;
Section 3 presents the data management and flow within the
platform and the user interaction. Finally, conclusions and
future work are presented in Section 4.
2 Platform Development
The main motivation to work on this platform is the difficulty
to geolocate places and addresses stored in a relational
database, that are ambiguous or with partial information. An
address with missing components of the descriptive
normalized location elements may produce that a geocoder
algorithm misunderstands individual values and fails to find
the correct geographic location. For example, a street number
Volunteering assistance to online geocoding services through a
distributed knowledge solution
José Pablo Gómez-Barrón
Universidad Politécnica de
Madrid
Camino de la Arboleda s/n
Km 7 de la Carretera de
Valencia
28031 Madrid, Spain
jpablo.gomezb@gmail.com
Miguel Ángel Manso
Universidad Politécnica de
Madrid
Camino de la Arboleda s/n
Km 7 de la Carretera de
Valencia
28031 Madrid, Spain
m.manso@upm.es
Ramón Alcarria
Universidad Politécnica de
Madrid
Camino de la Arboleda s/n
Km 7 de la Carretera de
Valencia
28031 Madrid, Spain
ramon.alcarria@upm.es
Abstract
Geocoding process of unstructured or poor quality location addresses requires human supervision in order to obtain valuable data. Current
availability of geocoding web-service technologies, enabling the deployment of collaborative applications and the existence of volunteering
communities, has motivated the proposal of a platform to generate geocoding collaborative tasks relying on the available solutions in order
to get accurate results in short-term. In this work we present the design and development of a tool that facilitates to volunteers the geocoding
process. We implement some strategies to give the volunteers elaborated information about web-service geocoding results and the capacity
to propose other positions different than those suggested. All the information is registered in database model to enable later analysis or
accuracy studies.
Keywords: Geocoding task, web services, VGI, crowdsourcing, collaborative platform.

2. RICH-VGI: enRICHment of volunteered geographic information (VGI) - AGILE 2015
could be considered a postal code or a street name could be
interpreted as neighborhood or a locality. As a result, specific
elements of a query to an online geocoding service will be
incomplete and geocoded outputs may differ. To guarantee
the quality of a geographic final layer, human interation to
check and analyze the result is required.
In order to deal with these obstacles the platform uses a
crowdsourcing approach to facilitate a distributed online
participation by volunteered users to process an address,
assisting the online geocoding services. Supported by a web
mapping client to browse and explore contextual data and
reference geographic layers, the user makes a comparison of
the output locations and quality attributes provided by the
services to choose the best option. Finally the platform saves
the user-selected response to geocode the address, but also the
coordinates and quality info of each geocoder with the aim to
have a data model to enable further evaluation analysis of
geocoders quality with redundant answers of the
collaborators.
Based on the identified problem and presented objectives
the platform implements two main components with some
secondary or support elements. Figure 1 shows the model that
corresponds with the platform capabilities.
The first main component is a management and user profile
area where the users can create a geocoding task or register in
a created task to participate. The second component is where
the geocoding task takes place. This user interface relies on a
short random list with the locations or addresses as input, and
a map that enables the user to examine the different geocoders
outputs.
The remainder platform support components are the landing
page where project is described, presenting the objectives,
utility and use instructions, and also the processes to register
and sign into the platform since all tasks, actions and
displayed data are linked to a particular user.
The platform is primary build on Python and JavaScript
languages. We use Django high-level Python web framework
to facilitate a clean design and a more organized application
structure, with server-side Python models and defined
functions to process POST and AJAX requests made by the
HTML/JavaScript client. Moreover, the Django framework
template language facilitates the use of variables inside
template tags to easy pass output values and context data to
render them on the web client. Also simplifies the platform
security with the integration of authentication, registration and
account management of the platform users by installing the
“django-allauth” application within the project.
All data of users, tasks and task rewards are managed using
Django models, a Python class that gives automatically
generated database-access API. The models that contain fields
and behaviours of the data are related to a single database
table implemented in PostgreSQL.
The server-side function related to process an address input
to be geocoded, imports and employs the “Python Geocoder
API” library. This Python wrapper client supports most
popular geocoding web services and converts the different
responses between each other into a consistent and unified
JSON response. In our platform, Google, Bing, Here and
OSM (Nominatim) providers are enabled and can be requested
through this library to send the geocoders output to the client.
Figure 1: Crowd-geocoding platform component model

3. RICH-VGI: enRICHment of volunteered geographic information (VGI) - AGILE 2015
Regarding address and task result querying and storing, the
platform use CartoDB geospatial database that internally is
based on PostgreSQL and PostGIS to manage the geographic
outputs. The client uses the SQL JavaScript API to obtain
random addresses with the constraint that the user can process
these addresses only once, so the SELECT query considers the
login user id. To insert user-selected option and each geocoder
output on CartoDB results table, the Python client API for
CartoDB SQL service was used. This library supports OAuth
2.0 open standard, and the platform server-side scripts use it
in a transparent way to the final user.
Finally the web client utilizes some JavaScript libraries like
JQuery and Bootstrap for easier web development and
Leaflet.js and Mapbox.js for the provision of interactive web
maps. These maps are composed of reference base layers and
overlay data generated with geographic coordinates given by
the geocoder services.
3 Data Management and User Interaction
3.1 Platform-User Interaction
User interaction with the platform components is described in
Figure 2 and the geocoding task process is related with the
user interface elements presented in Figure 3.
Figure 2: Flowchart with main processes.
After user registration and login, users are directed to the
user profile and task management area. Inside this area the
user can administrate their information as a task creator or
collaborator and begin tasks. Some actions enabled in this
area are: create, explore and select tasks, upload and
download address data, explore user and task progress and
statistics, and manage rewards and communications with
collaborators. Also a geocoding task can be edited or
customized with specific data entries related to a project
particular needs like the selection of geocoder providers,
predefined geocoder area results filters, number of task
answers needed for an address to be considered completed, or
number of null task answers to consider an address invalid.
To begin the assisted geocoding process the user has to
select an address or location from the listed options on the left
panel area of Figure 3 and make a request to the server in
order to geocode the selected input. Then, on center and right
panels (Figure 3), respectively, the user explores and interacts
with the locations on the map and compares the data quality
attributes provided by the geocoding library. Based on these
data and contextual information obtained by toggling the base
map layers the user chooses the geocoding service that
considers more accurate or suggests a new point in the map as
geocoding output.
3.2 Data Management
In addition to the geocoded location three parameters are
stored in the database: accuracy, quality and confidence. An
example for accuracy values is the method used to calculate
the location; If the result was approximated, interpolated or is
a geometric center of a line or polygon area, or if corresponds
to a precise correct match as a postal address. Regarding to
the quality value, it represents the output match level or
granularity of the match, related with a location entity type
like street address or house number, route or street,
intersection, neighbourhood, postal code, etc.
The third parameter to evaluate the geocoder output quality,
confidence, is obtained from the OpenCage API calculation
method that uses the bounding box data response from the
each API to create a confidence range between 0 and 10. The
confidence is calculated by measuring the distance in
kilometres between the South West and North East corners of
each resulted associated bounding box; smaller distances
represent a high confidence while larger distances represent a
lower confidence. In the geocoding platform, this parameter
can help users to select the best output, as a normalized score
result and comparable between the different geocoding
services.

4. RICH-VGI: enRICHment of volunteered geographic information (VGI) - AGILE 2015
The platform benefits from the user evaluation tasks to
compare the performance between the geocoding services
storing all their information. The data model designed for this
purpose is presented in Figure 4.
4 Conclusions and Future Work
We present a crowdsourced collaborative approach to deal
with an actual problem in the use of online geocoding
services. Our approach facilitates the user interaction to
control and evaluate the accuracy of geocoded outputs relying
in the amount of collaborators reviewing results and the
combination of diverse reference sources to increase data
availability.
With the task results database, users and researchers can
generate descriptive statistics, make a comparative evaluation
or data quality assessment using common geocoder metrics
like completeness, positional accuracy against base line data
and similarity between services.
As future work, we will enable the possibility for the user to
correct or modify the input address text to reduce the
ambiguity of the entry to the geocoding process and iterate the
process with the accumulative text editions. Also enable users,
in the task creation, to indicate the entity or geographic
feature that corresponds to the input address to geocode (e.g
swimming pools), hence the user can identify the best
accurate geocoder related to the searched physical object in
the base map.
Figure 4: General data model.
Figure 3: Crowd-geocoding platform user interface

5. RICH-VGI: enRICHment of volunteered geographic information (VGI) - AGILE 2015
References
[1] H. A. Karimi, M. Durcik, and W. Rasdorf, “Evaluation
of uncertainties associated with geocoding techniques,”
Comput. Civ. Infrastruct. Eng., vol. 19, no. 3, pp. 170–
185, 2004.
[2] D. W. Goldberg, J. P. Wilson, and C. a. Knoblock,
“From Text to Geographic Coordinates: The Current
State of Geocoding,” URISA J., vol. 19, pp. 33–46,
2007.
[3] P. A. Zandbergen, “A comparison of address point,
parcel and street geocoding techniques,” Comput.
Environ. Urban Syst., vol. 32, no. 3, pp. 214–232, 2008.
[4] D. Roongpiboonsopit and H. a. Karimi, “Comparative
evaluation and analysis of online geocoding services,”
Int. J. Geogr. Inf. Sci., vol. 24, no. April 2015, pp. 1081–
1100, 2010.
[5] R. Hudson-Smith, A., Batty, M., Crooks, A., Milton,
“Mapping for the masses: accessing web 2.0 through
crowdsourcing.,” Soc. Sci. Comput. Rev., vol. 27 (4), pp.
524–538, 2009.
[6] C. Heipke, “Crowdsourcing geospatial data,” ISPRS J.
Photogramm. Remote Sens., vol. 65, no. 6, pp. 550–557,
Nov. 2010.
[7] M. F. Goodchild, “Citizens as sensors: the world of
volunteered geography,” GeoJournal, vol. 69, no. 4, pp.
211–221, Nov. 2007.