=Paper=
{{Paper
|id=None
|storemode=property
|title=MapXplore: Linked Data in the App Store
|pdfUrl=https://ceur-ws.org/Vol-905/Veres_COLD2012.pdf
|volume=Vol-905
|dblpUrl=https://dblp.org/rec/conf/semweb/Veres12
}}
==MapXplore: Linked Data in the App Store==
https://ceur-ws.org/Vol-905/Veres_COLD2012.pdf
MapXplore:
Linked Data in the App Store
Csaba Veres
University in Bergen, Norway
csaba@csabaveres.net
Abstract. MapXplore is an attempt to build a mainstream, useful, and easy to use
application that uses linked data at its core. As such, it will be one of very few such
apps on the Apple, or for that matter any of the mobile app stores. The purpose of the
application is to allow users to browse any part of the globe, and identify points of
interest from DBPedia articles. They can then drill down into traditional as well as
linked data sources to get a comprehensive view of the points of interest. We note
some difficulties in working with current linked data resources, and suggest some
methods to help pave a future rich in popular, easy to use mobile semantic
applications. By using these guidelines we aim to keep refining MapXplore to make it
a showcase application for the power of linked data.
Keywords. Mobile, linked data, iPad, iOS, dbpedia, wikipedia, tourism
1 Introduction
Mobile computing is perhaps one of the most ubiquitous and fastest growing
technological innovations in recent years. Only sixteen years ago, [1] predicted that
the then rapidly expanding technology of cellular communication would eventually
give rise to a future in which millions of people would carry portable computing
devices which would not need a physical network connection. Their vision soon
became a reality as wireless technologies continued to improve, which prompted the
release of innovative hardware platforms like the Apple inc.® iPhone® in 2007,
which was perhaps the first device to bring a general, user friendly computing
platform into widespread use. Apple inc.® subsequently opened the “App store”
where native applications for their iOS platform could be downloaded for free or for a
relatively low price (compared to PC based applications). The App store opened on
July 10, 2008, and as of June 2012, reported over 30 billion downloads from the 365
million iOS devices sold through March 30, 2012.1 Following Apples’s phenomenal
success many other manufacturers opened their own application stores, and Google
inc.® released their Android® platform which copied many of Apples’s usability
features. Together, the two platforms dominate the mobile platform landscape [2].
It is possible that the runaway success of smartphone devices is due in no
small part to the native apps. The story goes that Steve Jobs was initially opposed to
third party applications on the iPhone, preferring instead Web2.0 and Ajax apps that
run on the Safari engine built into the iPhone. However, web apps never became
popular, so in 2008 developers were given access to the native SDK, and the wildly
successful app store opened in conjunction with the release of the new iPhone 3G.2
The usability guru Jakob Nielsen has shown that native apps on mobile devices are far
superior to web apps, and the main motivation for preferring the development of web
apps is more based on strategy and finance than usability.3
The rise in popularity of mobile “smart phones” has become most staggering
in just the last couple of years. A Nielsen report from Q3 2011 shows that 44 percent
of U.S. mobile subscribers own a smartphone device as of Q3 2011, compared to 18
percent just two years ago, and in the 25-34 age group the figure is 64 percent.
Moreover, the number of smartphone subscribers using the mobile Internet has grown
45 percent since 2010 [2].
The ubiquity of smartphone devices has been used to advantage by app
developers who think of new ways to exploit location based mobile use. A prominent
example is the so called SOCIAL + LOCAL + MOBILE or “So-Lo-Mo” trend, which
1 http://events.apple.com.edgesuite.net/126pihbedvcoihbefvbhjkbvsefbg/
event/index.html
2 http://www.cultofmac.com/125180/steve-jobs-was-originally-dead-set-
against-third-party-apps-for-the-iphone/
3 http://www.useit.com/alertbox/mobile-sites-apps.html
is a term for marketing efforts to target their information through a localized, social
platform which is sensitive to users’ environmental context. [2] reports that in the
United States, 87 percent of people use deal-of-the-day sites like Groupon4 which
offers current deals in the user’s immediate vicinity.
Apart from these lucrative marketing efforts, feature rich mobile devices
have also given rise to a large number of applications that help users explore their
immediate environment for information, entertainment, food, companionship and
many other opportunities. The major search engines have responded to the need for
people to know “where am I and what is around me?” with services like Google
Places.5
In addition to the real world applications, there has also been a huge amount
of academic work on the topic of Mobile Computing and Context-aware systems e.g.
[3]. We will not review these in any depth, because the focus of this paper is the
experience of developing a usable, mobile semantic application.
The application presented in this paper is predicated on the feeling that, since
the integration of information is a defining characteristic of linked data, there should
be a large number of linked data applications for contextually rich information
integration in the various application stores. Sadly, this is not the case. The “mobile
semantics” blog 6 is collating a list of iOS and Android apps which claim to make use
of semantic technologies in some form. The number of semantic apps at this moment
is vanishingly small, and most of them provide functionality that is esoteric, or does
not provide clear advantages over competing applications.
In spite of the somewhat disheartening status quo on the mobile app stores,
we believe that mobile semantic apps have a bright future. In order to achieve
success, however, future apps need to fulfill a number of requirements. The author of
this paper is working on defining a set of desirable characteristics that a successful
mobile semantic application must possess.7 Currently there are four conditions:
• The semantics should help rather than hinder the user. The app should not simply
present 100 possible links for the user to follow.
• The application should present some clear advantages over non semantic
versions. It should be able to do some clever and useful things that simply
cannot be done by competing apps of similar functionality, without
semantics. The semantics should make it the go to app in its domain.
4 http://www.groupon.com/
5 http://www.google.com/places/
6 http://mobilesemantics.blogspot.no/2012/05/mobile-semantic-apps-where-
are-they.html
7 http://mobilesemantics.blogspot.no/2012/05/mobile-semantic-apps-where-
are-they.html
• It should be usable. Nobody (almost nobody) wants to type SPARQL queries into
a 4 inch touchscreen!
• The semantics should be non trivial.
There are currently two prominent applications which fulfill these
requirements. The first is Siri, Apple’s personal assistant for iOS.8 Siri is able to carry
out simple inferences on users’ questions, in some domains. This is accomplished by,
what Tom Gruber calls it, “semantic snap to grid”.9 By modelling the user’s behavior
in a task oriented, contextually constrained information space, Siri is able to make a
guess about intended meanings of questions.
A second application is news360˚ which is a news aggregator that uses
semantic technologies for selecting and ranking related news stories in a personalized
manner.10 Its use of semantics is non trivial and completely transparent, and its
selection of related stories is impressive.
MapXplore aims to be a user friendly app which introduces users to the
world of linked data, and is a first step in the attempt to bring semantics into popular
use. In this first release, the focus is on fulfilling requirements two and three, to create
a user friendly application which showcases the usefulness of linked data. The
semantics is currently limited to simple semantic categorization. The application also
contains an implementation of Google analytics allowing us to track its use and
ascertain which features are used, so we can see whether or not the features are found
useful.
In the remainder of the paper we describe a use case which prompted the
development of MapXplore. We look at some existing apps with related functionality,
and discuss why a linked data application might be needed to better fit the
requirements. Then we report on challenges as well as successes of developing the
application with linked data.
2 Use case and Existing Applications
The use case centers around the idea of exploring points of interest (POIs) in
a particular geographical area. This could be in situ, as in the case of a tourist who is
in a foreign location, or in exploration mode where a person can navigate to any part
of the globe on a map and explore points of interest. The application should highlight
these points, and provide some means for selecting potential points for further
investigation. That is, there should be some system by which the POIs are grouped
8 http://www.apple.com/iphone/features/siri-faq.html
9 http://vimeo.com/5424527
10 http://news360app.com/
into coherent categories. Once a POI is selected, there should be ways to find
additional information about it. The information should be comprehensive, and
include visual and factual information. The POIs should be of broad interest, not
focusing on just a few targeted categories like “restaurants” and “hotels”. Above all
the application should be easy to use, performing quickly and providing information
in an intuitive fashion.
There are existing applications which provide broad, location based
information content rather than specialized services such as hotel bookings, for
example.
Localscope11 is a “Location Browser for your iPhone”, which uses the
current GPS location to retrieve information from a large number of services.
However, even though the breadth of data is large, the information is not integrated in
any sense beyond the fact that they share a common geospatial bounding box. For
example, Wikipedia and Google Places might have information about the same
resource in a given area, but there is no connection between the two information
sources. Altogether the experience feels somewhat disconnected because each service
returns an overlapping set of independent results. However, it does provide a very
comprehensive search interface, and users can search by address, facility (gas station,
wifi, cafe), and so on, and returns informative results.
Roamz12 is an application which focuses more on the socially oriented
services. It provides filtered content from Facebook, Foursquare, Instagram and
Twitter, based on a personal profile. While the breadth of information on Roamz is
smaller than on Localscope, it is more connected because clicking on a result will
retrieve more results about the same entity. However, none of the results are designed
to give factual information as they are all from the social media. The intent of the app
is to facilitate spontaneous discovery based on catchy photographs and comments
rather than informative content.
DBPedia mobile [4], [5] is an application built to explore the “geospatial
semantic web”. It is based on DBPedia entries that have a geospatial property, so the
articles can be retrieved based on a location of interest. The POIs are retrieved from a
SPARQL endpoint and placed on the map with an icon corresponding to their YAGO
category[6]. Selecting any icon will pop up a Fresnel based Linked Data browser[7],
which integrates information from various linked data sources and gives access to
photographs from ‘flickr(™) wrappr’.13 The application is a brilliant demonstrator
which includes the ability to filter results by their properties, as well as execute full
SPARQL queries. Unfortunately the author has had very little success in interacting
with the service. Its interface is quite rough and can be difficult to navigate, partly due
to the fact that it is not a native app. The application fails to retrieve any POIs around
11 http://itunes.apple.com/us/app/localscope/id409869453?mt=8
12 http://itunes.apple.com/app/roamz/id459343660
13 http://www4.wiwiss.fu-berlin.de/flickrwrappr/
the author’s location in Bergen, Norway, forcing him to chose one of the few
predefined locations. Browsing POIs with the Fresnel browser is patchy and
unreliable at best, since results are often not retrieved. In fact, this complete reliance
on an external service for integrating information is the single biggest usability issue,
since a failure in that service renders the entire application useless.
WorldExplorer14 is an application for iOS as well as a version for OSX,
which also shows POIs from WikiPedia, but in a more user friendly manner. For a
given location it shows Wikipedia POIs with a thumbnail, which can be ranked by
distance from current location or by rating. The bottom left panel shows Google
maps, which can optionally be viewed in streetview. On the right one can read the
Wikipedia article about the selected POI. There is also current weather information.
The application is very highly rated in both the iOS and OSX app stores.
MapXplore can be regarded as a combination of DBPedia mobile and
WorldExplorer. It is an easy to use and attractive application which includes most of
the existing functionality of DBPedia mobile, with an immersive browsing interface.
It includes many more information sources about each resource than WorldExplorer
by virtue of linked data sources, and includes a much richer interface for selection of
locations of interest, by virtue of semantic classification.
3 MapXplore Overview
MapXplore is a native iPad application. Its main window is a map view
which zooms to the user’s current location when the app launches (fig. 1). It retrieves
surrounding POIs from DBPedia, within a certain distance of the current location.
Currently this is set at 2KM., but the distance is user configurable. The reason it is set
at such a close distance is that some places, for example the city of London contain
very many POIs per square kilometer, and the application makes no attempt to filter
the results before presentation. The defined distance returns a reasonable number of
results for most locations, and the user has the option to adjust the range after they see
the returned results. The location can also be selected in the search bar, or by moving
the map to any location and “dropping the pin” with a long press on the iPad screen.
14 http://itunes.apple.com/us/app/world-explorer/id381581095?mt=8
Fig. 1. MapXplore main window
The left table view shows the WikiPedia categories that describe the POIs,
with the labels ordered by the number of POIs in each category. The interesting side
effect of this is that the “character” of a city is quickly apparent just by looking at the
categories. For example Bergen is clearly famous for its geography and in particular
its mountains.
POIs can be selected in various ways, including by choosing a category
which reveals the contained POIs, which can then be selected. In fig. 2 the user has
selected Bergen Museum, which triggers a popup with the available data sources. The
sources are broadly listed under “Pictures” or “Information”. The information is
retrieved individually from the services, so there is no single point of failure, even
though individual services might intermittently fail to return a result.
Fig. 2. Data sources for the selected resource
3.1 Pictures
There are currently five image sources. The first three are purely location
based, and the remaining two are semantic services in which the picture is linked to
the resource itself. This mixture serves two purposes. First it provides a richer
experience to the user who can investigate the resource from more perspectives.
Second, it enables a quantitative evaluation of the coverage and quality of the
semantic services.
• Google streetview image API.15 provides static images taken outward in a radial
direction from the specified location. MapXplore retrieves four images at 0,
90, 180 and 270 degrees from the central location and displays them in a
popup image viewer.
• Google Maps JavaScript API v316 embeds Google maps with full streetview
functionality in an embedded popup browser window.
• Flickr17 the original photo sharing site, which was partly responsible for the
inception of Web2.0. Users upload photographs and add freeform tags. Many
of the photos are taken with mobile phones and contain GPS location data.
The Flickr API has several options for returning photographs, and in
15 https://developers.google.com/maps/documentation/streetview/
16 https://developers.google.com/maps/documentation/javascript/streetview
17 http://www.flickr.com/
MapXplore we retrieve photos ranked by “interestingeness” in a 1Km. radius
around the location of the resource itself. No effort is made to ensure that the
photograph is actually a depiction of the resource itself.
• flickr TM wrappr18 provides images from Flickr, but attempts to return only
highly relevant images by matching DBPedia metadata against flickr tags
and titles. The pictures will therefore most often be of the resource itself
rather than something else in the vicinity. The images are interlinked with
DBpedia using the foaf:depiction relationship type.
• Ookaboo19 “is a collection of free pictures, indexed by precise terms from the
semantic web.” The semantic API allows retrieval by RDF Freebase
identifiers, DBpedia Resource identifiers, or Wikipedia pages. MapXplore
retrieves photos with the DBPedia identifier.
3.2 Information
The three information sources contain linked data which the user can explore
to enrich their understanding of the domain.
• DBPedia links provide is a web view of the set of RDF triples about a given
resource, for example http://dbpedia.org/page/Bergen. The web view
contains clickable links), so users can follow the information graph in any
way they choose. While this may not be the most user friendly way to
browse the data, we feel that it is not a bad first experiment. Many of the
links have clear and descriptive names, so users should have no trouble
finding interesting facts like populationMetro, postalCode, octHighC,
octLowC, and so on.
• Sindice20 is a service which extracts and indexes semantic markup in the form of
RDF, RDFa, Microformats or Microdata from web sites, in the form of RDF
triples. It provides several ways to search and query, including free text
searches, exact matches, predicates, triples, and even has a SPARQL
endpoint. It returns web sites matching the query, as well as the RDF triples
which were scraped from the web sites. MapXplore performs a simple query
with the name of the resource, and presents the Sindice search results in a
popup browser window.
• Freebase21 is “an open, Creative Commons licensed repository of structured data
of almost 23 million entities”.
18 http://www4.wiwiss.fu-berlin.de/flickrwrappr/
19 http://about.ookaboo.com/a/about/what_is_ookaboo
20 http://sindice.com/
21 http://www.freebase.com/
Altogether, the information sources provide a rich mashup of information
about a particular resource. But the mashup is a collection of semantic and non
semantic data. We think this is important because it lowers the barrier for entry by
providing familiar components alongside a new set of novel features. Our expectation
is that the familiar components will encourage users to explore the unfamiliar
territory, and this clam will be tested once usage data becomes available.
The data sources are a mixture of transparent and not so transparent
semantics. In particular, DBpedia and Sindice present an interface that is less than
ideal, since it requires some investment of effort to navigate. One of the major design
plans for future releases is to streamline the presentation and exploration of linked
data from these sources, in particular by making them contextually sensitive to
respond to individual user needs and profiles.
4 Technical issues and challenges
One of the major advantages of building the application with semantic
technologies is that the retrieved POIs can be modified without any changes to the
application itself, since queries are made via SPARQL to a selected triplestore. The
query simply retrieves the labels and category names for all resources within the
specified geolocation. If the content of the triplestore is modified by the inclusion of
any resource whatsoever, it will show up on MapXplore.
An unfortunate problem, however, is in the reliability of the data sources.
During the development process it became painfully obvious that the server at http://
dbpedia.org/sparql would not reliably return results. Many times it would simply time
out without a response. Our solution was to set up a second server with replicated
data, and switch over if dbpedia.org failed to respond within five seconds.
A second problem was the response speed, which could be slow because the
SPARQL queries used FILTER clauses to retrieve just the resources in the specified
range. Some triplestores provide a special mechanisms for geo data. For example
Virtuoso has some built in functions (e.g. bif:st_intersects) that can be used to filter by
location.22 Unfortunately one can’t rely on using these functions in general because
they have to be specially set up by the administrator. For example, a second Virtuoso
instance at http://live.dbpedia.org/sparql does NOT support the functions. The impact
of this issue is exacerbated by the previous problem, because the queries have to be
written with sufficient generality to work with potentially any triplestore.
Another solution might be to consider an alternative query method
for retrieving the Wikipedia articles, and generating a DBPedia identifier from that.
While the Wikipedia API does not support query by location, there are several third
party solutions. The most straightforward is WikiLocation 23. However this only
22 http://docs.openlinksw.com/virtuoso/rdfsparqlgeospat.html
23 http://wikilocation.org/
returns the Wikipedia article identifier, so an extra lookup would be required to gather
information about semantic category, etc.
A more flexible option is a new “big data” service at Infochimps, who have
an API for Wikipedia data.24 This API is far richer than WikiLocation. The problem
with the API, however, is that it is not possible to retrieve selected category
information. Infochimps appear to have inserted their own categorization, but this
does not seem as useful or flexible as the categories retrieved through SPARQL from
DBPedia and its various mappings.
5 Preliminary evaluation
As previously mentioned, MapXplore contains a Google analytics
implementation, and we have some preliminary results based on one month’s
availability. The application has been downloaded in 37 countries, and received a total
of more than 1000 page views.
The application is an excellent platform for comparing information sources
linked from a unique identifier. It is easy to gather data about the coverage and
content of the different sources. Freebase, for example, appears to have fairly good
coverage of DBPedia resources, but the information does not seem to contribute
significantly: much of the text is simply a copy of the Wikipedia article, but some of
the structured information is useful. Flickr(™) wrappr returns photographs of the
resource itself more reliably than Flickr itself, so the semantic algorithm is good.
However, its coverage can be patchy. Similarly for Oookaboo. The overall subjective
impression is that the linked data provide accurate additional information, but their
coverage is patchy.
The preliminary results seem to support these observations. The most
frequently used features of the app are listed in descending order: retrieving and
searching for more POIs, static street view, Wikipedia, Oookaboo, flickr wrappr, street
view, DBPedia, Freebase, Flickr, Sindice. Users seem to favour the traditional
information sources, but are very keen to try the semantically enhanced versions.
Oookaboo and flickr wrappr are more than twice as often consulted than Flickr itself.
The more complex DBPedia browser is more popular than Freebase. On the other
hand the Sindice interface appears too confusing at the moment. Clearly future
development will involve streamlining the DBPedia and Sindice user experience.
6 Conclusion
The combination of mobile and semantic technologies is surely going to
break into the popular market in the coming years, especially as the availability and
reliability of open data expands. We proposed a set of development criteria to help
24 http://www.infochimps.com/datasets/wikipedia-articles
focus designers’ attention in making successful applications. We don’t want to
deceive anyone that it is going to be easy. In spite of the massive growth of open
linked data and consequent exposure in mainstream technology circles, it can still be
difficult to find reliable, extensive data for interesting applications that can capture the
public imagination.
MapXplore is a first attempt at making a linked data explorer tool that is both
feature rich as well as immersive and easy to use on a mobile platform. It uses a
combination of semantic and traditional technologies to gather information about geo
located points of interest. We hope that this combination will make a well received
and useful addition to many people’s app libraries. Subsequent releases will look to
additional data sources, enhanced usability of the information contained in the
sources, and a bigger role for semantic integration.
7 References
[1] T. Imielinski and H. F. Korth, “Introduction to Mobile Computing,” in The Kluwer
International Series in Engineering and Computer Science, vol. 353, no. 1, T.
Imielinski and H. F. Korth, Eds. Boston, MA: Springer US, 1996, pp. 1–43.
[2] Nielsen, “State of the Media: The Mobile Media Report Q3 2011,” nielsen, Dec.
2011.
[3] J. Hong, E. Suh, and S. J. Kim, “Context-aware systems: A literature review and
classification,” Expert Systems with Applications, vol. 36, no. 4, pp. 8509–
8522, 2009.
[4] C. Becker and C. Bizer, “DBpedia Mobile: A Location-Enabled Linked Data
Browser,” presented at the 1st Workshop about Linked Data on the Web
(LDOW2008), Beijing, China, April 2008., 2008.
[5] C. Becker and C. Bizer, “Exploring the geospatial semantic web with dbpedia
mobile,” Web Semantics: Science, 2009.
[6] F. Suchanek, G. Kasneci, and G. Weikum, “Yago: A Large Ontology from
Wikipedia and Wordnet,” Web Semantics: Science, Services and Agents on the
World Wide Web, vol. 6, pp. 203–217, 2008.
[7] E. Pietriga, C. Bizer, D. Karger, and R. Lee, “Fresnel: a browser-independent
presentation vocabulary for RDF,” presented at the ISWC'06: Proceedings of
the 5th international conference on The Semantic Web, 2006.