Web-based Evidence Excavation to Explore the Authenticity of Local Events
Ryong Lee Daisuke Kitayama Kazutoshi Sumiya
School of Human Science and Environment
University of Hyogo, Japan
Outline• Motivating Problem
– Authentic Local Information Search on the Web– Evidence Search for Local Events
• Credibility Layer in LBS Platform– Toward Credible Next Generation LBS
• Our Approaches:– Web-based Evidence Excavation, similar to the work of Archaeologists– Computational Model to validate Credibility
• Conclusion
QuickTime¢‚∞˙ æ–√‡ «ÿ¡¶±‚∞°
¿Ã ±◊∏≤¿ª ∫∏±‚ ¿ß«ÿ « ø‰«’¥œ¥Ÿ.
Motivating Problem: Authentic Local Information Search
• Is it true?
: Tom’s visit in our small town? Surprise!!
• How can we believe it?
: Can we find the evidence also on the web?
• We need ‘Evidence Search’ functions for local events on the web.
Evidence Search for Local Events
User Usage Procedure:1. A user browses a page about real-
world events
2. The user selects a part of interests on the browsing page.
3. The system detects the selected texts and generate a query to do evidence search.
4. Perform evidence similarity search
5. List up the found evidence list by degree of their credibility
A User Interface for Evidence Search
QuickTime¢‚∞˙ æ–√‡ «ÿ¡¶±‚∞°
¿Ã ±◊∏≤¿ª ∫∏±‚ ¿ß«ÿ « ø‰«’¥œ¥Ÿ.
Credible LBS Platform
Map the Webthe Web
LBS App.LBS App.
Map the Webthe Web
Credible LBS App.Credible LBS App.
CredibilityValidationCredibilityValidation
Map control Evidence control
Two Info. Sources
static dynamic
Credible LBS Platform
Map the Webthe Web
LBS App.LBS App.
Map the Webthe Web
Credible LBS App.Credible LBS App.
Map control Evidence control
CredibilityValidationCredibilityValidation
Problem: Web-based Evidence Excavation
1. Web pages can always change and disappear- It is important to keep track of the changes. --> Web Archive
2. Huge size makes it impractical to process queries in real-time. - pre-indexing/clustering in spatio-temporal dimensions- query : (time,space)-> event lists (as evidence)
3. Quality/Objectiveness Control:- credibility computation method: trustworthiness and authority
Evidence Excavation Process
EventEvent
WebWeb• Event: (time, space, vestige)
• Spatial Dimension Analysis: - Geocoding: geowords--> positions on the map
• Temporal Dimension Analysis: - datetime recognition in regular expression
• Event Clustering: - by a condition of ‘similar time periods AND near places AND similar keyword vectors’
What’s the difference between Archaeologists and us ?
Archaeologists Web-based Evidence Search
Target
Source
How-to
Periods
humanity’s material remains: buildings, art and even bodies
web documents or media
treasures or materials to know history evidence about real-world events
events in past age events in past, current and future
from the earth-born time to now? from the 1990 to now (very short period, relatively)
A painstaking digging or whipping?
web archiving and analysis
Physical Size
At most, volume of the earth?
sum of the storage size?
Imaginary Size
#(old historic stories) At least, #(the Web pages)exponentially increasing
Credibility Computation:Trustworthiness and Authority
Trustworthiness: degree of worthiness of confidence
Authority: degree of power cited or appealed to as an expert
Trustworthinessof events
Trustworthinessof events
Authorityof info. sources
Authorityof info. sources
based on Definition from the Merriam-Webster Online Dictionary
Both degrees should be changed dynamically.
Don’t believe the words of your boss !• Milgram Experiment*1: measured the willingness of study participants to obey an
authority figure who instructed them to perform acts that conflicted with their personal conscience. (Stanly Milgram, Yale Univ. 1963)
*1: Milgram, Stanley. (1974), Obedience to Authority; An Experimental View. Harpercollins.
Experimental Setting:- Experimenter(E) orders the the teacher(T) to ask questions and give penalties to a Learner (L), if L cannot solve them. - T believes that painful electric shocks go to L from 45V to 450V(!) as a penalty.- L in reality has no shocks and just scream.- L and T are separated in different two rooms.- T should continue giving problems to L and giving penalties by the order of E, even though the screaming continues.
How many percent of participants pushed the 450V killing Voltage to the students giving up their conscience?
command
question/penalty answer/
scream65% of participants pushed to the 450V!!
Dynamic CredibilityTo be more objective and independent on specific authorities,trustworthiness and authority should be changed dynamically.
QuickTime¢‚∞˙ æ–√‡ «ÿ¡¶±‚∞°
¿Ã ±◊∏≤¿ª ∫∏±‚ ¿ß«ÿ « ø‰«’¥œ¥Ÿ.
QuickTime¢‚∞˙ æ–√‡ «ÿ¡¶±‚∞°
¿Ã ±◊∏≤¿ª ∫∏±‚ ¿ß«ÿ « ø‰«’¥œ¥Ÿ.
Trust depends on Authorities.Good trustful data is being referred to by many good authorities.
Authority depends on Trust.Good authority has many trustful evidence.
based on HITS
Exampleevents
e1
e2
e3
e4
e5
sitess1s2s3s4s5s6s7s8s9s10
QuickTime¢‚∞˙ æ–√‡ «ÿ¡¶±‚∞°
¿Ã ±◊∏≤¿ª ∫∏±‚ ¿ß«ÿ « ø‰«’¥œ¥Ÿ.
starting from self-evaluation
starting from same authority
QuickTime¢‚∞˙ æ–√‡ «ÿ¡¶±‚∞°
¿Ã ±◊∏≤¿ª ∫∏±‚ ¿ß«ÿ « ø‰«’¥œ¥Ÿ.
QuickTime¢‚∞˙ æ–√‡ «ÿ¡¶±‚∞°
¿Ã ±◊∏≤¿ª ∫∏±‚ ¿ß«ÿ « ø‰«’¥œ¥Ÿ.
Conclusion• A framework of Evidence Search for Local Events on the
web was introduced.
• For objective and dynamic credibility estimation of the web contents, an interdependent evaluation method based on trustworthiness and authority was presented.
• Based on these computational credibility validation, we believe that Next Gen. LBS will be much more CREDIBLE !!
Credible LBS Platform
Map the Webthe Web
LBS App.LBS App.
Map the Webthe Web
Credible LBS App.Credible LBS App.
Map control Evidence control
CredibilityValidationCredibilityValidation
Future work
Problem: Map Consistency Control• Map is a key platform to integrate with various information.• Maps out-of-dated can cause inconsistency problems later.• Cartographers are hard to follow up map-update requests so fast.• But someone on the Web knows map changes and writes them over a
personal blog page.• If we can find them and are possible to translate the web contexts into
an map update contexts(delete/add/change,etc.)….
Challenging Task: Map Update using Web Contents Analysis
Future work
Map Update using Web Contents Analysisdisappear move appear
AB
BA
A
B
C
BX
XB
B
A
C
co(A,B) co(A,B) co(A,C) co(X,B)
identify (X)
t t t
t1
t2weaker
strong weakstrong
weaker stronger
stronger
C
C
strong
strong
co(B,C)
Future work