geotagging photographs by sanjay rana
DESCRIPTION
Presentation proposes a formal approach to represent and analyse tag clouds, using the principles of cartography and graph theory.TRANSCRIPT
GeoTagging Photographs |Sanjay
Rana
Contents
• Introduction
• Current Streams of Research
• Cartography of Tag Clouds• Topology of Tag Clouds
• Future Directions
“A picture is worth thousand words.”Fred Barnard,
An advertising manager in the early 1920s
Introduction
•GeoTagging is the process of assigning geographical information metadata, called geo-tags, to various media e.g. photographs, video, websites etc.
• A tag is a term/label/keyword that describes some property of an object. It probably - a) originated from a Scandinavian/German word meaning point/prong/spike/prickle/thorn, b) is a cognate of tack (Source: etymonline.com).
Introduction…
Tags•lyon •france •2005 •night •october •geotagged
Machine tags (2) •geo:lat=45.75 •geo:lon=4.82
Source: http://www.flickr.com/photos/nodomain/57506095/
GeoTags
Introduction…• Some observations:
– The primary purpose of GeoTagging is to assign suitable identifiers to a photograph so that searching in a large catalogue of photographs can produce relevant results quickly.
– Therefore as a concept, it is basically same as the focus of large body of research on Geographical Metadata e.g. FGDC in the last 40 years, and that of broader Archival Science of over several hundreds of years and perhaps even thousands of years.
– So, what's different about GeoTagging?
Introduction…• GeoTagging is different
because:
– Crowd-sourced / Volunteered Geographical Information
– Strongly linked to online community (NeoGeography)
– Free-form text with very nominal formatting structure except for the geographical coordinates
– Inconsistent quality
Introduction…
• Tags are valuable because they :Help to generate User-
Oriented and User-Relevant indexing of information in databases i.e. Ego-Centric Design & Mental Map.
Reveal fascinating insights into the sociological behaviour & semantics of participants – Vernacular Geography, Folksonomy, Narrative Geography, Storytelling, Locative Media.
Introduction…
• Aim of this presentation is to propose:– Cartographical Approach set in a – Graph-Theoretic Framework that can be used primarily by :– System Designers &– Exploratory Users for displaying and analysing relationships
between all Tags i.e. not only geo-tags.• Disclaimer: In progress ideas..
Contents…
• Introduction
• Current Streams of Research
• Cartography of Tag Clouds• Topology of Tag Clouds
• Future Directions
Current Streams of ResearchBroadly three streams of research
1. Creation of Tags2. Visualisation of Tags3. Analysis of Tags
Current Streams of Research…• Creation of Tags
– Largely technological Topics:
– Hardware related research:
• Robust integration of GPS adapters, cameras, handheld devices, desktop devices etc to create a seamless approach.
– Software related research:
• Automated association of tags using geographical and image parameters, cool mash-ups, storage standards (EXIF, ICBM, RDF, GeoBloggers, GeoHash, FlickrFly…).
Current Streams of Research…• Visualisation of Tags
– Ever increasing ways to display the tags and relationships between tags.
• Active interest from researchers from various disciplines including Information Visualisation, HCI, Graphic Arts, Social Sciences,…
• ManyEyes Project by IBM.– Most basic approach involves a static 2D plot where
tag’s font size is proportional to its count/frequency in the database. Example
– Advanced visualisations are animated and interactive and generally display more complex relationships between tags. Example.
– Seen as an viable alternative to navigation of large databases.
– However, no widely accepted guidelines or approach to generate a tag cloud. For example,
Current Streams of Research…• Which of the following colour
classification is more informative?
Current Streams of Research…• Which of the following label styles is
more informative?
Current Streams of Research…• Analysis of Tags
– Implicit ways of organising the tags into meaningful clusters i.e. unlike the explicit ways, which involve user-led exploratory visualisation
• Ordinary frequency distributions.• Semantic Similarity and Semantic Relatedness
measures– Various statistical, probability etc. type measures for
measuring relationships between tags - kind of like Google’s ranking. For example,
– Showing only few representative photos of a city from millions of photos.
– Recognising that “Lift” and “Elevator” are similar
– Understanding social networking and sociological profiling of users.
Contents…
• Introduction
• Current Streams of Research
• Cartography of Tag Clouds• Topology of Tag Clouds
• Future Directions
Cartography of Tag Clouds
• Hypothesis:– Tag Clouds are essentially a sort of
Topological Cartograms. Example– Tag’s text label is analogous to the
geographical feature or symbol.
• Implications of the hypothesis:– Availability of vast bank of design
guidelines and usability studies in Cartography to represent Tag Clouds. For example,
Cartography of Tag Clouds…• The optimal number of thematic classes
in a map was found to be 7+2 (Miller)
Application of Cartographic Principles
Cartography of Tag Clouds…• Other (but not all) useful Cartography
Concepts:– Techniques of Cartographic Abstraction and
Generalisation.– Visual Variables for both Static and Dynamic
Maps e.g.:– Choice of Colour Classification– Choice of Font Face/Size/Weight or using Symbols– Brushing to show relationships between two
representations.– Reducing Clutter and Redundancy
– Multi-modal e.g. Haptic visualisation, sounds, etc.
• Cartography Cube - A benchmark to evaluate a Tag Cloud representation?
Cartography of Tag Clouds…
User Audience
Interaction
Data Relation
s
Public
Private
Low
High
Unknown
Known
Contents…
• Introduction
• Current Streams of Research
• Cartography of Tag Clouds• Topology of Tag Clouds
• Future Directions
Topology of Tag Clouds
• A Tag Cloud can be represented as an:– Undirected– Non-Planar Graph G=(V,E) where
V = {Tag1, Tag2, Tag3,…}; E = Some connectivity basis e.g. co-occurrence, semantic relatedness etc.
• Several interesting graph-theoretic measures can be derived to visualise and analyse the structure of the tag clouds.
• For example,
Topology of Tag Clouds…
• What is the least number of tags and photos that will represent the contents of all the photographs in the database?
• Is the Tag Cloud graph a Scale-Free Network i.e. are there any patterns in the database of photographs?
Sand (60%) Aquarium (22%)
Beach (40%)
Palm Tree (2%)
Yacht (25%)
0
600
0
20
600
90
Frequency
Co-occurrence
10
0
Contents…
• Introduction
• Current Streams of Research
• Cartography of Tag Clouds• Topology of Tag Clouds
• Future Directions
Future Directions…
– Technological• Instrument accuracy/portability/affordability,
data formats, System scalability (massive information).
• Metadata Standards.
– Conceptual• Exploration and Application of Geo-Tags• Risks to the Netizens
– Misinformation - Queen storming out or walking in.– Google search collapse by incorrect tagging of all
links to be bad.– Privacy - Revelation of personal preferences.
• In summary, it is proposed that the visualisation and analysis of Tag Clouds can be formalised by using the principles of Cartography and Graph Theory.
• Thank you!
Simple Tag Cloud
Source: ManyEyes Project
Keywords For Photographs Of Dubai
Advanced Tag Cloud
Elastic Tag Maps by Moritz Stefaner
Source: well-formed-data.net
Cartogram
Number of Internet Users in 2002
Source: WorldMapper.Org
Visual Variables
Source: How Maps Work (MacEachren, 2004)
Dynamic Maps
Source: GeoVista, University of Pennsylvania
Scale-Free Network
Is it the case that there are a few tags that exist in a lot of photographs while most photographs have dissimilar tags – Scale Free Network and Small-World Networks?
Misinformation
How to Lie with Maps (Monmonier 1996) = How to lie with Photographs or How to lie with geo-tags…?