Semantic Analytics on Social Networks: Experiences in Addressing the Problem of Conflict of Interest Detection : Semantic Analytics on Social Networks: Experiences in Addressing the Problem of Conflict of Interest Detection World Wide Web 2006 Conference
May 23-27, Edinburgh, Scotland, UK This work is funded by NSF-ITR-IDM Award#0325464 titled '‘SemDIS: Discovering Complex Relationships in the Semantic Web’ and partially by ARDA Boanerges Aleman-Meza1, Meenakshi Nagarajan1,
Cartic Ramakrishnan1, Li Ding2, Pranam Kolari2,
Amit P. Sheth1, I. Budak Arpinar1, Anupam Joshi2, Tim Finin2 1LSDIS lab
Computer Science
University of Georgia, USA 2Department of Computer Science and Electrical Engineering2
University of Maryland, Baltimore County, USA
Outline : Outline Application scenario: Conflict of Interest
Dataset: FOAF Social Networks + DBLP Collaborative Network
Describe experiences on building this type of Semantic Web Application
Conflict of Interest (COI) : Conflict of Interest (COI) Situation(s) that may bias a decision
Why it is important to detect COI?
for transparency in circumstances such as
contract allocation, IPOs, corporate law, and
peer-review of scientific research papers or proposals
How to detect Conflict of Interest?
connecting the dots
Scenario for COI Detection : Scenario for COI Detection Peer-Review: assignment of papers with the least potential COI
Our scenario is restricted to detecting COI only
(not paper assignment)
Current conference management systems:
Program Committee declares possible COI
Automatic detection by (syntactic) matching of email or names, but it fails in some cases
i.e., Halaschek Halaschek-Wiener
Conflict of Interest : Conflict of Interest Verma Sheth Miller Aleman-M. Thomas Arpinar Should Arpinar review Verma’s paper?
Social Networks : Social Networks Facilitate use case for detection of COI
But, data is typically not openly available
Example: LinkedIn.com for IT professionals
Our Pick: public, real-world data
FOAF, Friend of a Friend
DBLP bibliography
underlying collaboration network
Covering traditional and semantic web data
Our Experiences: Multi-step Process : Our Experiences: Multi-step Process Building Semantic Web Applications involves a multi-step process consisting of:
Obtaining high-quality data
Data preparation
Metadata and ontology representation
Querying / inference techniques
Visualization
Evaluation
Our Experiences: Multi-step Process : Our Experiences: Multi-step Process Building Semantic Web Applications requires:
Obtaining high-quality data
DBLP, FOAF data
FOAF – Friend of a Friend : FOAF – Friend of a Friend Representative of Semantic Web data
Our FOAF dataset was collected using Swoogle (swoogle.umbc.edu)
Started from 207K Person entities (49K files)
After some data cleaning: 66K person entities
After additional filtering, total number of Person entities used: 21K
i.e., keep all ‘edu/ac’
DBLP ( ) : DBLP ( ) Bibliography database of CS publications
Representative of (semi-)structured data
We focused on 38K (out of over 400K authors)
authors in Semantic Web area
arguably more likely to have a FOAF profile
DBLP has an underlying collaboration network
co-authorship relationships
Combined Dataset of FOAF+DBLP : Combined Dataset of FOAF+DBLP 37K people from DBLP
21K people from FOAF
300K relationships between entities
Our Experiences: Multi-step Process : Our Experiences: Multi-step Process Building Semantic Web Applications requires:
Data preparation
Our goal: Merging person entities that appear both in DBLP and FOAF
Person Entities from two Sources : Goal: harness the value of relationships across both datasets
Requires merging/fusing of entities Person Entities from two Sources
Merging Person Entities : Merging Person Entities We adapted a recent method for entity reconciliation
- Dong et al. SIGMOD 2005
Relationships between entities are used for disambiguation
Presupposition: some coauthors also appear listed as (foaf) friends
With specific relationship weights
Propagation of disambiguation results
Syntactic matches : DBLP Researcher Amit P. Sheth UGA Marek Rusinkiewicz Steefen Staab John Miller http://www.informatik.uni-trier.de/~ley
/db/indices/a-tree/s/Sheth:Amit_P=.html Dblp homepage http://lsdis.cs.uga.edu/~amit/ coauthors homepage label FOAF Person Carole Goble Ramesh Jain John A. Miller Amit Sheth Professor 9c1dfd993ad7d1852e80ef8c87fac30e10776c0c http://www.semagix.com
http://lsdis.cs.uga.edu http://lsdis.cs.uga.edu/~amit affiliation friends Workplace
homepage label title homepage Syntactic matches mbox_shasum
… with Attribute Weights : DBLP Researcher Amit P. Sheth UGA Marek Rusinkiewicz Steefen Staab John Miller http://www.informatik.uni-trier.de/~ley
/db/indices/a-tree/s/Sheth:Amit_P=.html Dblp homepage http://lsdis.cs.uga.edu/~amit/ coauthors homepage label FOAF Person Carole Goble Ramesh Jain John A. Miller Amit Sheth Professor 9c1dfd993ad7d1852e80ef8c87fac30e10776c0c http://www.semagix.com
http://lsdis.cs.uga.edu http://lsdis.cs.uga.edu/~amit affiliation friends Workplace
homepage label title homepage … with Attribute Weights mbox_shasum The uniqueness property of the Mail box and homepage values give those attributes more weight
Relationships with other Entities : DBLP Researcher Amit P. Sheth UGA Marek Rusinkiewicz Steefen Staab John Miller http://www.informatik.uni-trier.de/~ley
/db/indices/a-tree/s/Sheth:Amit_P=.html Dblp homepage http://lsdis.cs.uga.edu/~amit/ coauthors homepage label FOAF Person Carole Goble Ramesh Jain John A. Miller Amit Sheth Professor 9c1dfd993ad7d1852e80ef8c87fac30e10776c0c http://www.semagix.com
http://lsdis.cs.uga.edu http://lsdis.cs.uga.edu/~amit affiliation friends Workplace
homepage label title homepage Relationships with other Entities mbox_shasum A coauthor who is also listed as a friend
Propagating Disambiguation Decisions : DBLP Researcher Marek Rusinkiewicz Steefen Staab John Miller coauthors FOAF Person Carole Goble Ramesh Jain John A. Miller friends Propagating Disambiguation Decisions If John Miller and John A. Miller are found to be the same entity, there is more support for reconciliation of the entities Amit P. Sheth and Amit Sheth
based on the presupposition that some coauthors an also be listed as (foaf) friends
Results of Disambiguation Process : Results of Disambiguation Process Number of entity pairs compared: 42,433
Number of reconciled entity pairs: 633
(a sameAs relationship was established) 49 205 379 DBLP 38,015
Person
entities 21,307
Person
entities FOAF
Our Experiences: Multi-step Process : Our Experiences: Multi-step Process Building Semantic Web Applications requires:
Metadata and ontology representation
(How to represent the data)
Assigning weights to relationships : Assigning weights to relationships Weights represent collaboration strength
Two types of relationships (in our dataset)
‘knows’ in FOAF (directed)
‘co-author’ in DBLP (bidirectional)
Anna co-author Bob
Bob co-author Anna
Assigning weights to relationships : Assigning weights to relationships Weight assignment for FOAF knows Verma Sheth Miller Aleman-M. Thomas Arpinar FOAF ‘knows’ relationship
weighted with 0.5 (not symmetric)
Assigning weights to relationships : Assigning weights to relationships Weight assignment for co-author (DBLP)
#co-authored-publications / #publications
The weights of relationships were represented using Reification Sheth Oldham co-author co-author 1 / 124 1 / 1
Our Experiences: Multi-step Process : Our Experiences: Multi-step Process Building Semantic Web Applications requires:
Querying and inference techniques
Semantic Analytics for COI Detection : Semantic Analytics for COI Detection Semantic Analytics:
Go beyond text analytics
Exploiting semantics of data (“A. Joshi” is a Person)
Allow higher-level abstraction/processing
Beyond lexical and structural analysis
Explicit semantics allow analytical processing
such as semantic-association discovery/querying
COI - Connecting the dots : COI - Connecting the dots Query all paths between Persons A, B
using ρ operator: semantic associations query
Anyanwu & Sheth, WWW’2003
Only paths of up to length 3 are considered
Analytics on paths discovered between A,B
Goal: Measure Level of Conflict of Interest
Trivial Case: ‘Definite’ Conflict of Interest
Otherwise: High, Medium, Low ‘potential’ COI
Depending on direct or indirect relationships
Case 1: A and B are Directly Related : Case 1: A and B are Directly Related Path length 1
COI Level depends on weight of relationships
Sheth Oldham co-author co-author 1 / 124 1 / 1
Case 2: A and B are Indirectly Related : Case 2: A and B are Indirectly Related Path length 2
Verma Sheth Miller Aleman-M. Thomas Arpinar Number of co-authors
in common > 10 ?
If so,
then COI is: Medium Otherwise, depends on weight
Case 3: A and B are Indirectly Related : Case 3: A and B are Indirectly Related Path length 3
Verma Sheth Miller Aleman-M. Thomas Arpinar COI Level is set to: Low
(in most cases, it can be ignored) Doshi
Our Experiences: Multi-step Process : Our Experiences: Multi-step Process Building Semantic Web Applications requires:
Visualization
Visualization : Visualization Ontology-based approach enables providing ‘explanation’ of COI assessment
Understanding of results is facilitated by named-relationships
Our Experiences: Multi-step Process : Our Experiences: Multi-step Process Building Semantic Web Applications requires:
Evaluation
Evaluating COI Detection Results : Evaluating COI Detection Results Used a subset of papers and reviewers
from a previous WWW conference
Human verified COI cases
Validated well for cases where syntactic match would otherwise fail
We missed on very few cases where a COI level was not detected
Due to lack of information or outdated data
Examples of COI Detection : Examples of COI Detection Wolfgan Nejdl, Less Carr
Low level of potential COI
1 collaborator in common
(Paul De Bra co-authored once with Nejdl and once with Carr) Stefan Decker, Nicholas Gibbins
Medium level of potential COI
2 collaborators in common
(Decker and Motta co-authored in two occasions,
Decker and Brickley co-authored once,
Motta and Gibbins co-authored once,
Brickley and Motta never co-authored,
but Gibbins (foaf)-knows Brickley) Demo at http://lsdis.cs.uga.edu/projects/semdis/coi/ or, search for: coi semdis
Our Experiences: Multi-step Process : Our Experiences: Multi-step Process Building Semantic Web Applications involves a multi-step process consisting of:
Obtaining high-quality data
Data preparation
Metadata and ontology representation
Querying / inference techniques
Visualization
Evaluation
Evaluation : Evaluation Demo at http://lsdis.cs.uga.edu/projects/semdis/coi/ or, search for: coi semdis Underlined: Confious would have failed to detect COI
Our Experiences: Discussion : Our Experiences: Discussion What does the Semantic Web offer today?
(in terms of standards, techniques and tools)
Maturity of standards - RDF, OWL
Query languages: SPARQL
Other discovery techniques (for analytics)
such as path discovery and subgraph discovery
Commercial products gaining wider use
… Our Experiences: Discussion : … Our Experiences: Discussion What does it take to build Semantic Web applications today?
Significant work is required on certain tasks
such as entity disambiguation
We’re still on an early phase as far as realizing its value in a cost effective manner
But, there is increasing availability of:
data (i.e., life sciences), tools (i.e., Oracle’s RDF support), applications, etc
… Our Experiences: Discussion : … Our Experiences: Discussion How are things likely to improve in future?
Standardization of vocabularies is invaluable
such as in MeSH and FOAF; but also: microformats
We expect future availability/increase of
Analytical techniques used in applications
Larger variety of tools
Benchmarks
Improvements on data extraction, availability, etc
What do we demonstrate wrt SW : What do we demonstrate wrt SW We demonstrated what it takes to build a broad class of SW applications: “connecting the dots” involving heterogeneous data from multiple sources- examples of such apps:
Drug Discovery
Biological Pathways
Regulatory Compliance
Know your customer, anti-money laundering, Sarbanes-Oxley
Homeland/National Security
…..
Our Contributions : Our Contributions Bring together semantic + structured social networks
Semantic Analytics for Conflict of Interest Detection
Describe our experiences in the context of a class of Semantic Web Applications
Our app. for COI Detection is representative of such class
Data, demos, more publications at SemDis project web site, http://lsdis.cs.uga.edu/projects/semdis/ Thanks! Questions : Data, demos, more publications at SemDis project web site, http://lsdis.cs.uga.edu/projects/semdis/ Thanks! Questions
References : References Related SemDis Publications (LSDIS Lab - UGA)
B. Aleman-Meza, C. Halaschek-Wiener, I.B. Arpinar, C. Ramakrishnan, and A.P. Sheth: Ranking Complex Relationships on the Semantic Web, IEEE Internet Computing, 9(3):37-44
K. Anyanwu, A.P. Sheth, ρ-Queries: Enabling Querying for Semantic Associations on the Semantic Web, WWW’2003
C. Ramakrishnan, W.H. Milnor, M. Perry, A.P. Sheth, Discovering Informative Connection Subgraphs in Multi-relational Graphs, SIGKDD Explorations, 7(2):56-63
Related SemDis Publications (eBiquity Lab – UMBC)
L. Ding, T. Finin, A. Joshi, R. Pan, R.S. Cost, Y. Peng, P., Reddivari, V., Doshi, J. and Sachs, Swoogle: A Search and Metadata Engine for the Semantic Web, CIKM’2004
T. Finin, L. Ding, L., Zou, A. Joshi, Social Networking on the Semantic Web, The Learning Organization, 5(12):418-435
Other Related Publications
X. Dong, A. Halevy, J. Madahvan, Reference Reconciliation in Complex Information Spaces, SIGMOD’2005
B. Hammond, A.P. Sheth, K. Kochut, Semantic Enhancement Engine: A Modular Document Enhancement Platform for Semantic Applications over Heterogeneous Content, In Kashyap, V. and Shklar, L. eds. Real, World Semantic Web Applications, Ios Press Inc, 2002, 29-49
A.P. Sheth, I.B. Arpinar, and V. Kashyap, Relationships at the Heart of Semantic Web: Modeling, Discovering and Exploiting Complex Semantic Relationships, Enhancing the Power of the Internet Studies in Fuzziness and Soft Computing, (Nikravesh, Azvin, Yager, Zadeh, eds.)
A.P. Sheth, Enterprise Applications of Semantic Web: The Sweet Spot of Risk and Compliance, In IFIP International Conference on Industrial Applications of Semantic Web, Jyväskylä, Finland, 2005
A.P. Sheth, From Semantic Search & Integration to Analytics, In Dagstuhl Seminar: Semantic Interoperability and Integration, IBFI, Schloss Dagstuhl, Germany, 2005
A.P. Sheth, C. Ramakrishnan, C. Thomas, Semantics for the Semantic Web: The Implicit, the Formal and the Powerful, International Journal on Semantic Web Information Systems 1(1):1-18, 2005