COPLINK A National Model for Law Enforcement Information

COPLINK: A National Model for Law Enforcement Information Sharing and Knowledge Management Acknowledgement: NSF, DARPA, NIJ, TPD, PPD Hsinchun Chen, Ph. D. Director, Artificial Intelligence Lab/Hoffman e. Commerce Lab, University of Arizona

Presentation Overview • IT Challenges in Law Enforcement • COPLINK Origins • COPLINK Solution • COPLINK Benefits • COPLINK Demonstration • COPLINK Future Directions

Current Dilemma in Law Enforcement • Mobile, generalist criminal element • Public and political pressure to perform • Resource constraints in terms of time, money and people • Information systems integration challenges

IT Scorecard - The Good • Increasing awareness of importance of technology tools to enhance productivity • Information is being transferred into digital form • IT has put data in the hands of many more employees • Agencies have collected a tremendous amount of data

IT Scorecard – The Challenges • Specialized policing needs have slowed solutions • Access to capital has been a limiting factor • Resulting information systems are incompatible • Critical crime related data can be cumbersome to access • Stores of data may be grossly underutilized • Regional information sharing initiatives have been limited

Intra Agency Problem -Tucson Example Stand Alone Databases Records Management System (RMS) Gang Database Mugshots Database

Inter Agency Problem -Statewide Example Incompatibility for Information Sharing* County Sheriff Systems Police Department Systems State Police Systems * Exception being statewide criminal history information

COPLINK: Origins, Solution, Benefits

COPLINK Progression 1990 Artificial Intelligence Lab (AI Lab) founded at UA 1992 NSF CISE funding; CIA Russian computing analysis 1994 NSF Digital Library Initiative (DLI-1) funding; core indexing and analysis technologies 1997 NIJ COPLINK funding; Web-enabled data warehousing 1998 NSF DLI-2 funding, DOD DARPA funding; spider/agent technologies 1999 DOD and NSA Intelink; NIH Medical informatics 2000 NSF Digital Government funding, NIJ AGILE interoperability funding

COPLINK & AI Lab In the Press July/1994 Science Semantic retrieval technologies Feb/1996 IEEE COMP NSF digital library research Oct/1997 Science Web retrieval and analysis May/1998 Gov Tech COPLINK review May/1999 IEEE COMP NSF digital library research Jun/1999 TECHbeat COPLINK: Database Detective Sep/1999 NY Times Cyber content mapping Feb/2000 JASIS Digital library research Apr/2000 Civic. com Changing the Rules of the Game Apr/2001 Am. Police Tracking criminals is suddenly Beat easier

NATIONAL INSTITUTE OF JUSTICE Office of Science and Technology Research Technology Development Division AGILE: Interoperability and NIJ/OST What have we done for interoperability lately? 1995 NLECTC-Rocky Mt opens (main focus area is interoperability) 1996 BORTAC 1996 Info. Tech 1997 COPLINK 1998 NIJ publishes Research Report “State and Local Law Enforcement Wireless Communications and Interoperability: A Quantitative Analysis” 1998 AAG’s interoperability video “Why Can’t We Talk” is released 1998 Consolidation of all interoperability projects under a focused program - AGILE (Source: NIJ AGILE)

NATIONAL INSTITUTE OF JUSTICE Office of Science and Technology Research Technology Development Division COPLINK est. 1997 GOAL: Interjurisdictional information sharing that allows sophisticated analysis & data mining. METHODOLOGY: 1. Uses data warehousing to facilitate information sharing. 2. Has a software program that analyzes data and relationships. 3. Uses "concept space, " a knowledge-mining tool that identifies the relationships between objects such as people, vehicles, organizations, locations, weapons, or crimes. FIRST DOCUMENTED SUCCESS: Tucson police agreed to help a Federal agency with a homicide case. Sole information known was a tip that the suspect (no name known) had a sister (no name known) living in Tucson who several years ago had been assaulted by her boyfriend (whose name was known). In less than a minute, the system returned the woman's name and the name of her brother. CURRENT TEST SITE: Tucson and Phoenix, Arizona (Source: NIJ AGILE)

COPLINK Executive Summary • COPLINK is a problem solving technology using a distributed database design • COPLINK resides within the agency’s secure intranet so control of information is optimized • COPLINK connects most separate databases which provides accessibility • COPLINK goes much further to analyze data to uncover hidden crime leads • COPLINK allows one stop links to other local, state, and federal databases

Traditional Warehouses Defined • Developed for data consolidation and speed of access • Used a central repository approach that enabled control, security and privacy • Challenged by static data issues and data loss issues • Derived value for users through a large number of contributors

The COPLINK Distributed Design • Distributed intranet/extranet based, web-enabled design optimizes control, security and privacy • Data mirror approach safeguards against the overload of contributing mission critical systems • Automated porting and refresh algorithms overcome data re-entry, static data, and data loss issues • Flexible, user centric design provides value for single agency needs, and is scalable to accommodate regional problem solving • Enhanced data analysis power via relationship identification and data/text mining

Web-enabled

Interoperability

Interoperability

Interoperability

Interoperability

Scalable

Scalable

Scalable

Scalable

Secure and Private • Users hit node only – User queries never penetrates existing databases • Secure Intranet/Extranet solutions – Firewall, IP address, user password, search audit trails – Encrypted and compressed HTTPS/IP based over dedicated line or VPN • Privacy is controlled locally – No intelligence information ported into the node – Agency determined sharing via policy-guided solution

Software Components

COPLINK Connect Consolidating & Sharing Information promotes problem solving and collaboration Records Management Systems (RMS) Gang Database Mugshots Database

COPLINK Detect Consolidated information enables targeted problem solving via powerful investigative and data/text mining applications

COPLINK Connect: Examples

COPLINK Connect Person Search

COPLINK Connect

COPLINK Connect Location Search

COPLINK Connect

COPLINK Connect Vehicle Search

COPLINK Connect

COPLINK Detect: Example

COPLINK Detect The Crime: An eight-year old female is abducted from a neighborhood park. The only witnesses are three other children. They provide vague physical descriptions of two male suspects who forced the victim into a green 2 -door car. One witness heard one suspect call the other “Dave. ”

COPLINK Detect • In this type of crime, time is of the essence. • Curb-side access to the COPLINK Detect program can provide investigative leads quickly. • We want to explore any relationships between “Dave” and a green 2 -door.

COPLINK Detect • Our first step is to find all persons whose name begins with “Dav. ” • “Dave” could be nickname for “David, ” “Davidson, ” “Davis, ” etc.

COPLINK Detect

COPLINK Detect • The results list 41 persons that match the search criteria. • This is too many to manually sort through to see if they are associated with a green car. • So we will add the entire “results set” to the relationship window.

COPLINK Detect

COPLINK Detect Now we will want to search for “green 2 -door” cars.

COPLINK Detect

COPLINK Detect • The results set for “green 2 -door” is five vehicles. • Since we don’t know which vehicle we are seeking, we add all “green 2 -door” vehicles to the relationships window.

COPLINK Detect

COPLINK Detect Now we are ready to look for any relationships between all the “Dav” and all the “green 2 -door” results sets.

COPLINK Detect

COPLINK Detect • We now see that “Davis N. Whitehead” and “Damien Fiscus” were together in a green 1990 Buick Skylark, Arizona Lic. MZA-262. • The incident was a “suspicious vehicle call. ” • We can now see all incidents involving Davis Whitehead.

COPLINK Detect

COPLINK Detect • Davis Whitehead is listed as a suspect only in the “suspicious vehicle” incident. • Now we need to look at Damien Fiscus.

COPLINK Detect

COPLINK Detect • We now examine the second individual in the “suspicious vehicle” incident – Damien Fiscus.

COPLINK Detect

COPLINK Detect • We see that Damien Fiscus is involved in two additional incidents: – A vehicle collision report – A sexual assault case

COPLINK Detect

COPLINK Detect Our Results: • We now know that Damien Fiscus was previously arrested for sexual assault. • We have addresses and pictures of both Fiscus and Whitehead. • We have a description and license plate number for Fiscus’ vehicle. • This doesn’t mean we have solved the crime, but these individuals are good investigative leads.

COPLINK Connect Usability Measures • Effectiveness – Impact on Job Performance & Productivity – Accuracy of system – Effectiveness of information displayed • Ease of Use – Effort to learn and use system – Navigation through the system • Efficiency – Interface design – Speed

Interview Results • General Themes -Speed: “ 100% quicker”, “Saves time”, “Get information quicker” -Ease of use: “A lot easier to use”, “I could use it without training” -Interface: “Less steps to get “information”, “Flexible to organize [sort]” -Information: “A lot more information than RMS”, “Enter less information, the more you get”

COPLINK Detect Evaluation • Explore impact that knowledge management techniques can have in the investigative arena • Questions asked: – How will software be used? For what tasks? – Can investigative tool lead to increased case solvability? – Where should the future development focus be?

Solvability Improvements • Comparison of Time Spent per Search using RMS versus Detect • 15 Actual Case Searches • Average Saving of 31 minutes per search • 65% time save

COPLINK Benefits • Enhances organizations problem solving capabilities in terms of crime, fear, and disorder • Improves case solvability through greater access to information and knowledge • Provides tool to improve officer safety • Results in better decision making and subsequent employee job satisfaction • Is a cost-effective way to link and analyze data within existing stovepipe systems • Has minimal training requirements

COPLINK Deployment Status • TPD: COPLINK Connect operational since January, 2001 (400+ users, 24/7) • TPD: COPLINK Detect under deployment (30+ users); July 2001 • Arizona COPLINK Connect/Detect: October 2001 • Arizona COPLINK Agent: May 2002

COPLINK Future Directions

COPLINK NSF Team • AI Lab: Dr. Hsinchun Chen, Director, AI Lab/COPLINK Center Dr. Homa Atabakhsh, Associate Director, COPLINK Center Drs. Zeng, Fenstermacher, Zhao Ph. D Students Rosie Hauck Michael Chau Jennifer Xue Haidong Bi Jing Zhang Masters Students Stanford Pugsley Michael Huang Lihua Cao Simon Chen Shan Fu Yang Xiang Xue Wei Undergrads Yi Qin • Tucson Police Department: Sgt. Jennifer Schroeder • Yintak Lam Detective Tim Petersen Phoenix Police Department: Joseph T. Hindman, Computer Services Bureau Administrator

Research Areas • Visualization: revealing crime patterns and associations • Text mining: narrative analysis • Agent and collaboration: push, spider, and wireless • COPLINK sociology: a profession under transformation

Visualization

Hyperbolic Tree • Basics of hyperbolic tree – Focus + Context – Distorted view of tree • Algorithm Tree drawing algorithm on hyperbolic plane – The angle of each sub-tree is as big as the angle of its parent – Exponentially more space is available with increasing distance from center • Reference – “A Focus+Context Technique Based on Hyperbolic Geometry for Visualizing Large Hierarchies. ” • http: //www. acm. org/sigchi/chi 95/Electronic/documnts/papers/jl_bdy. htm – Resource: Andreas Hadjiprocopis • http: //www. soi. city. ac. uk/~livantes/PROGRAMS/Hyperbolic. Tree/hyperbolic. html

New Features • Hierarchical tree and Hyperbolic tree views • Retrieve and display information dynamically • Use color to distinguish different data types • Filter information using checkboxes • Show weight of relationship between search term and resulting nodes through arc thickness • Display details using tool tip • Highlight duplicated information (nodes)

Text Mining

Text Mining: Narrative Analysis • Arizona Noun Phraser, AZNP (topic identification) • Concept Space (crime relationships) • Self-Organizing Map, SOM (text mining) • Entity Extraction (who/what/where) • Data Set and Preprocessing – Source: PPD narrative data (1998) – Selected crime type: Narcotics – Size: 1300 narcotic related cases (8 M) – Format: Plain text SGML – Selection of stop terms: Iterative improvement

Concept Space: Crime Relationships

SOM: Text Mining

Narrative Analysis — Entity Extraction • Entity Extraction: Combining computational linguistics and neural network techniques to extract conceptual entities from text, e. g. person, location. • The system consists of 3 components: – Boundary Identifier: AZNP • To parse input text into noun phrases. – Finite State Automata, FSA • To determine patterns of noun phrase and co-located words before and after the phrase. – Neural Network, Backprop • To apply machine learning on word patterns.

Narrative Analysis — Entity Extraction • Entity Extraction for General Text

Narrative Analysis — Entity Extraction • Applying to narrative documents – Characteristics of narratives • All texts are typed in upper case. • Many sentences are incomplete and ungrammatical. – Customizations • Program do not make use of case information. • Make use of specific patterns in narratives. • New terms are added to the lexicon. • Extracts 2 entity types: Person and Address.

Narrative Analysis — Entity Extraction • Entity Extraction for Narrative Document

Narrative Analysis — Entity Extraction • Results of Pilot Study – General Text • Precision: 89. 8% • Recall: 83. 9% – Narrative • Precision: 74. 9% • Recall: 52. 7% • Current development • Customize the program for narrative data to improve precision and recall rates. • Expand the system such that more entity types (e. g. , vehicles and narcotics) can be extracted.

Agent and Collaboration

COPLINK Patrol-Secure Wireless • Allows Patrol Officers to enhance their community expertise • Further promotes Officer safety through curbside knowledge • Laptop, PDA, and cell phone access

COPLINK Agent • Enables 24 hour police work through on and off shift monitoring of information Intelligent Spider Application • Provides advanced collaboration tools for interjurisdictional information sharing and cooperation

COPLINK Collaboration System • Objective – Develops an architecture for information sharing and collaboration in the law enforcement domain. • User Requirement Study – Interviews and focused group studies were conducted at the Tucson Police Department. • Functionalities desired but not currently available – Monitor data on an entity or a search query. – Locate sergeants/detectives in other units who work on related cases. – Share useful information for investigation. • Technical Requirements – Security is of utmost importance. – Confidentiality/Anonymity: People may not be willing to share data.

COPLINK Collaboration System Architecture

COPLINK Collaboration System • Functions of System Modules – Information Access and Monitoring • Retrieves and combines useful data from distributed data sources. • Schedules periodic monitoring of data sources using agents – Security and Confidentiality Management • Ensures secure access of data. • Allows users to specify the desired levels of sharing and confidentiality. – Collaboration • Applies data mining techniques. • Recommends data sources to user based on one’s information needs. • Identifies detectives working on similar cases.

COPLINK Sociology

Interacting with the LE Community • User-centered design: two full-time TPD officers/detectives • Focused, staged user studies: a sociology team • Quick prototyping and user feedback • TPD user briefings: 30+ management and user group briefings (7 IT Assistant Chiefs) • Arizona/regional partner briefings: 15+ regional partner meetings • National/regional NIJ/DOJ and LE meetings: 10+ LE IT meetings • Annual COPLINK Center Workshop: 1/25 -26, 2001 • Linking LE communities: Arizona, CA, TX, Utah, Nevada, Michigan, etc.

COPLINK Sociology Issues • Building trust: (LE, technologist, sociologists); (intra-inter agencies) • Understanding police culture • A profession under transformation (gun vs. laptop) • Reward and training • Employee turn-over and organizational memory: retaining knowledge • Time sensitive and mission critical: agent to the rescue!

COPLINK: Lessons Learned • Aim high, but walk first. • Technologies change, be adaptive. • Build one, but for all. • Solve their problems – user centered design. • Cool, but is it useful? • Early and consistent involvement of users, administrators, and IT personnel. • Technology adoption in organizational context. • Communication with your program officers. • (Product – Prototype) = $1. 5 M

For project information: http: //ai. bpa. arizona. edu hchen@bpa. arizona. edu