Semantic Web Solves Semantic Web Solves CrimesCrimes

Semantic Web Solves Semantic Web Solves CrimesCrimes

9 Feb 20069 Feb 2006Gregory FairnakGregory Fairnak

ConsultantConsultant

Speaker Bio• Consultant to Northrop Grumman Corp.• System Architect for Law Enforcement

Information Exchange (LInX)• Technical Lead State of Texas Fusion Center• Activities related to Business Development• Over 15 years systems integration experience

primarily in healthcare, public safety and homeland defense

• Office 954-783-0907 gfairnak@comcast.net

Introduction• Throughout the nation, law

enforcement agencies and systems integrators are striving to facilitate the flow of information to help prevent and solve crimes.

• Many large “data sharing” projects– If every officer knew what every officer

knows?• Is there a place for the semantic web?

SWANS 2005• Semantic Web Applications for National Security

– Mr. Tim Berners-Lee Director World Wide Web Consortium

• “Stop arguing what we call things”• “Context is what is really important”

– Deborah McGuinness, Stanford University senior research scientist

Artificial Intelligence Laboratory at Stanford University.

• Useful areas for application

• Natural fit to law enforcement records search

Semantic Web - Primer• Smarter Web• Organize Data into Knowledge• More than just keyword match• Software agents roam page to page

carrying out tasks for the user• Resource Description Framework or

RDF is the organizing mechanism

Problem areas addressed by Semantic Web

• Integrating Multiple Data Sources• Semantic Drilling Down• Statements About Statements• Translation• Smarter Search

– Source: McGuiness and Deans SWANS 4/17/05

Our Problem• Finding records across one or more databases

where there are multiple correct ways to describe and document what happened

• Law Enforcement Records Management Systems– A database of incidents and reports– Original purpose to generate crime statistics– New purpose generate leads and solve crimes

• Each system has own local dialect, lookup table • Data sharing architecture two predominate

patterns

One Approach

LHP

BSO

METRO

SOPD

WarehouseQuery

Query fieldsmappings

White, Male, Victim

Another Approach

LHP

BSO

METRO

SOPD

Query

White, Male, Victim

Our Problem (cont.)• User interface requires officer training

– multiple ways entering search/query – requires exact data entry – Do not allow for typos, synonyms, concept

searches like “robbery” or categories– Show me all, dark SUV’s, pickup trucks, red sports

cars

• National Institute for Justice Grant Request, – “Public safety officers and analysts need a

simplified and consistent interface”

• Missed Records

Our Problem (cont.)• Data

– Some fields are mapped to a standard (Race, Sex, Hair Color, Eye-Color)

– Other fields such as Offense Descriptions, Make and Model of Vehicles or Pawned Items are not

– Multiple ways to say the same thing

031 Robbery 1st Deg

ROBB 1

Robbery

ROBBERY 2

031 Robbery 2nd Deg

1201 Robbery

1202 Robbery

Vision• A detective is assigned a recent liquor store

robbery• Happened anywhere else, part of a series? • According to witnesses, the suspect wore a ski

mask, held a sawed off shot gun, only grabbed large bills. They describe the suspect as white male, 6’ tall, with blue eyes.”

• A software agent roams from page to page collecting relevant records

Requirements• Fast implementation, data centric• Work with both approaches to

data sharing• Based on Standards• Bird’s eye view no missed records• One Concept could have multiple

expressions

Demonstration• Show original agency incident records

without preferred terms and categories• Explain the user interface• Show agency incident records with

preferred terms and categories– Search for Robbery– Search White, Tall >6ft, Male– Liquor and ski mask

Solution Description

LHP

BSO

METRO

SOPD

SeamarkServer

Query

Browser

JSPXRBRXPath

SQL to XML

XSLTRDF

Faceted Navigation• Organize information the way officer’s

think – in flexible categories – for easier navigation

• Organize search to support in the way witnesses think

• Create categories based on SUV’s, Sports cars, gun manufacturer

• Balanced Scorecard

Strengths of semantic integration

• Fast first project• Easy search• No missed records• Everything is accessible, “big picture”• Ontology development can grow over time• Ability to inherit from others• Human insight leveraged to the context

Lessons Learned• XSLT fast,

– not elegant, might not scale• No impact on production systems• Sometimes an unexpected values pops up• OWL, RDF or Spreadsheet

– <element name><alt value><pref value>– <Sex> <m> <MALE>– <Make> <Chevy><Chevrolet>

• OWL organizing concept layer

Next Steps• Identify most important

dimensions (user group)• Build out additional categories and

dimensions (using OWL)– Pawn, Vehicle and Offense Types