"EventMedia Live: Exploring Events Connections in Real-Time to Enhance Content" presented at the Semantic Web Challenge, Open Track, of the 11th International Semantic Web Conference, Boston, USA, November 2012
Presentation on how to chat with PDF using ChatGPT code interpreter
EventMedia Live: Exploring Events Connections in Real-Time to Enhance Content
1. EventMedia Live: Exploring Events
Connections in Real-Time to
Enhance Content
Houda Khrouf, Vuk Milicic and Raphaël Troncy
{khrouf,milicic,troncy}@eurecom.fr
@rtroncy
Semantic Web Challenge, ISWC 2012, Boston
2. Events through the web …
Event Directories
Allows opportunistic/serendipitous discovery
Limited social features and Information
incompleteness (low coverage, weak decision)
Media Platforms
Aids remembering and sharing past experiences
Lack of consistent events descriptions
Social Networks Attendance: popularity and friends' whereabouts,
attendees’ opinions.
Limited discovery mechanism: need to filter out
relevant information.
Explore semantic connections to create better overviews of events
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 2
3. Approach
Aggregate heterogeneous data sources
Unify into one data model using Semantic
Web technologies
Enrich with media, social data and
background knowledge
Identify meaningful and/or interesting
relationships between events, media, and
people
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 3
4. User Tasks and Challenges
Goals:
Discover PAST, PRESENT and FUTURE events
Live, relive and predict experiences through shared media
Enhance search and recommendations mechanisms
Support the decision making process
Challenges
Semantization of Data (events, media, people)
Reconciliation of sparse and highly heterogeneous data
User-centric design: discovering, deciding and recommending
Scalability, Caching and Optimization
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 4
5. Architecture
Heterogeneous data sources
Real-time Reconciliation
- Tag-based mapping
Data Crawling
RDF Conversion
- Instance matching
RSS Update
Web Applications
SPARQL
REST API (Elda)
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 5
6. Data Crawling
Various API specifications
Policy management
Requests chaining
Different input and output schemas
Time-consuming
A Restful Service leveraging the commonalities of API specifications
A Web dashboard to easily handle data collection and interlinking
http://eventmedia.eurecom.fr/dashboard/
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 6
7. Data Modeling – LODE ontology
LODE is a minimal model that encapsulates the factual properties of
events: What, Where, When and Who.
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 7
8. Real-Time Reconciliation
Precise time
agents
Cost & description
Media
Participants
Need for semantic connections between events, media, agents and locations
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 8
9. Events and Media connections
APIs
Machine tags
1.7 million images over
“lastfm:event=”
126.000 events
“upcoming:event=”
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 9
10. Live Update
Per week
• 1500 Photos
Tag = *:event • 130 events
Number of Items posted during the last days Number of Items posted per days of the week
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 10
11. Instance Matching
Events/Agents/Locations semantic connections
SPARQL-based Instance matching
Filtering by the issuance date
Challenges
Structural heterogeneity
Different representations of the same value
User-generated knowledge featuring human mistakes
Different granularities between data sources
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 11
12. Events connections
The events similarity is a mutual agreement of their factual properties
Based on top-k dependencies between properties
p1 p2 dependency
1st level
Minimal conditions to fetch
title1 title2 0.30
similar events using SPARQL place1 place2 0.28
title1 agent2 0.26
2nd level
Refine the results agent1 agent2 0.21
description1 title2 0.16
Evaluation: Precision: 96% and Recall: 94%
Ground truth of 300 matched instances
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 12
13. Agents & Locations connections
Agents connections (precision-oriented)
Internal reconciliation + DBpedia, Musicbrainz and BBC
Label matching + contextual filtering (Cosine)
Granularity? (1 to N mappings) Cosine inequality
Cosine (d, a+b) > Cosine (d ,a)
Locations connections (precision-oriented)
Internal reconciliation + DBpedia and Foursquare
It is straightforward thanks to the consistency of locations descriptions
Top-k Google search
Geographical distance
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 13
14. About the dataset
Event Agent Location Photos User
Last.fm 61,173 62,577 15,917 1;464,885 130,249
Upcoming 19, 276 15 10,269 308,148 3,932
Eventful 45,204 8,725 18,763 3,028 321
Total 125,650 71,313 44,948 1,760,787 159,263
DBpedia MusicBrainz Eventful
Last.Fm 13064 34247 2865
Eventful 2515 3616 -
SPARQL endpoint: http://eventmedia.eurecom.fr/sparql
Restful API : http://eventmedia.eurecom.fr/rest/event/
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 14
15. Web Application
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 15
16. Web Application
Challenges:
Modern single-page app on top of semantic technologies
Fast, user-friendly and responsive UI
Dealing with high variability of data
Technologies:
Backbone.js: MVC framework enables complex UI,
provides flexibility
Elda Linked Data API maps SPARQL queries to RESTful
URLs
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 16
17. What's the added value of SW?
Data model (RDF) + URIs for all objects +
decentralized semantics (re-use of ontologies) =
life MUCH easier for data integration
Re-use data of others …
DBpedia, MusicBrainz, BBC, Foursquare, etc.
… and expose our reconciled data (CKAN)
Toolkits:
Virtuoso: triple store + SPARQL endpoint + RDF publishing
Silk / Knofuss: several similarity metrics for instance matching
Elda (Linked Data API)
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 17
18. Semantic Web Challenge criteria
Attractive/Functional UI: fluid navigation, multimedia
Scalable application: 40 million triples and counting
Novelty: real-time events connections discovery
Use of dynamic data: social media, rss feeds
Cross languages, cross devices: phones, tablets
Commercial potential: ask the entrepreneurs here!
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 18
19. What's on this evening?
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 19
20. What's on this evening?
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 20
21. Demo: Live your event
http://eventmedia.eurecom.fr
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 21
22. Future Work
User interface: aggregated views of events, agents,
locations and users.
Opinionated content retrieval: social networks, online
newspapers, etc.
Identify Interests and provide recommendations
Analysis of social networks activities around real-world
events (behavioral patterns)
14/11/2012 Semantic Web Challenge - ISWC 2012 - Boston 22