N092-149 TITLE: Similarity Measures for Persona/Human Networks

TECHNOLOGY AREAS: Information Systems, Human Systems

ACQUISITION PROGRAM: PM Intel, MCSC POR and Actionable Intel Enabled by Persistent Surveillance

OBJECTIVE: To develop an application that enhances the identification of at-risk actors and/or networks using robust closeness or similarity metrics. Human persona and networks can be described in terms of their past behavior, current activities, and external forces influencing its behavior. Signatures of at-risk groups can be described in similar terms. The objective of the topic is to automate the detection of at-risk personas and human networks through N-dimensional clustering and comparison to individuals or groups considered to be at-risk.

DESCRIPTION: Signatures of persona and human networks can be described as N-dimensional term vectors or tensors. Examples of terms are proximity to an idea or goals, communications patterns, interactions with individuals or other networks, proximity to themes, proximity to places, observed behaviors, memberships, structure, or stability (1). Each of these terms may be multidimensional vectors made up of observable and latent variables. New attributes could be dynamic and include attributes such as the semantic distance between a network and other entities (events, places, and people). Classic social-cultural insights must also be modeled as external force attributes.

Research is needed to find signatures of interest in large data bases that are similar to a known at-risk persona or human network. A human network tensor, once defined and computed, is intended to be precise enough to allow the recognition of the same network in disparate data sources and for the recognition of closeness between one network and another. Comparisons and classification decisions can also be based on examination of the difference between the human network vector and its past state in time or space.

Similarity measures have been developed for many applications. Commonly used measures are Pearson�s correlation coefficient or Euclidean distances. Multidimensional scaling (MDS) uses similarity measures to produce a psychological space in which similarity is inversely related to distance (2). Feature-based similarity measures result in a feature-matching process whereby common features increase similarity but those unique to a set decrease the similarity metric (3). Topological methods are applied in fields such a semantics, and graph theory is widely used for assessing similarities in taxonomy. Researchers at Ohio State University have developed an "extrinsic" similarity measure to surmise the similarity of two genes by the similarity of their relation with other genes (4). The University of Michigan has developed a "behavior bounding" hierarchical model method to compare human behavior to computer agents. (5) Can some instantiation or combination of these measures be applicable to the persona/human network problem?

The challenge is to develop similarity metrics that account for persona and human network vectors with terms that are not normalized, may be sparse, may be unequally filled, and may be dynamic. Measures of closeness need to be computed as soon as data are collected. Potentially, a composite metric or confidence score may be a gradient or indicator that the persona or human network is moving towards at-risk behavior.

The Navy is interested in innovative R&D that involves technical risk. Proposed work should have technical and scientific merit. Creative solutions are desired.

PHASE I: Identify and define a persona or human network signature to be studied, and develop similarity measure (clustering) algorithms that account for non-normalized and incomplete terms in a signature. Provide a theoretical description of the measures and a final report that is presented at a technical conference.

PHASE II: Simulate data that provide observations of persona/human network signature components. Produce a prototype system that is capable of ingesting simulated signature data and provides similarity measures as outcomes. The prototype should be able to cluster large populations into related sub-populations using complex n-dimensional signatures. Test the similarity methodology using disparate sources of observations and variability in the signature terms (missing terms, ambiguity, latent variables). An initial capability should have a probability of correct classification of 90% and false alarm rate below 10%.

PHASE III: Produce a system capable of deployment and operational evaluation. The system should address threats to specific operational environment (e.g. attacks on roadway convoys). It should operate in a distributed SOA environment, handle multiple data streams, and provide explanation to a user as to why a network or persona has been classified as at-risk. Evaluation of accuracy of alerts and false alarm rates will be made by system operation in an exercise or field environment. Product success will be judged by military operators and transition assistance provided by SPAWAR Systems Center Pacific.

PRIVATE SECTOR COMMERCIAL POTENTIAL: Commercial applications of persona and human network comparisons may in support of marketing and consumer analysis to monitor consumer purchasing activity. In addition, political groups may be interested in categorizing and predicting influence in various demographic regions. Law enforcement agencies can apply these methods to identify gangs or criminal groups based on their interactions and behaviors. Health organizations, such as the CDC, can apply these techniques to the prediction of infectious disease spread through human network behavior modeling and comparison. Presently, there is a strong need to protect military and civilian personnel from terrorist attacks and human network classifications tools are needed. The systems should operate in a net-centric environment and provide reliable performance. Commercial value and cost savings is achieved by operation in a SOA with other applications.

REFERENCES:
1. Pentland, A. Automatic Mapping and Modeling of Human Networks. Physica A (2007).

2. Young, F. W. & Hamer, R. M. Theory and Applications of Multidimensional Scaling. Hillsdale, NJ: Erlbaum, 1994.

3. Tversky, A. Features of Similarity. Psychological Review, 84, 1977, pages 327-352.

4. Ucar, D., Altiparmak, F., Ferhatosmanoglu, H., & Parthasarathy S., Investigating the Use of Extrinsic Similarity Measures for Microarray Analysis, BIOKDD �07, San Jose, CA, August 12, 2007.

5. Wallace, S., & Laird, J. Behavior Bounding: Toward Effective Comparison of Agents & Humans, Proceedings of the Eighteenth International Joint Conference on Artificial Intelligence, 2003.

KEYWORDS: human networks, similarity measures, terrorist threats, cognitive science

** TOPIC AUTHOR (TPOC) **

DoD Notice:   Between April 20 and May 17, 2009, you may talk directly with the Topic Authors to ask technical questions about the topics. Their contact information is listed above. For reasons of competitive fairness, direct communication between proposers and topic authors is not allowed starting May 18, 2009, when DoD begins accepting proposals for this solicitation.

However, proposers may still submit written questions about solicitation topics through the DoD's SBIR/STTR Interactive Topic Information System (SITIS), in which the questioner and respondent remain anonymous and all questions and answers are posted electronically for general viewing until the solicitation closes. All proposers are advised to monitor SITIS (09.2 Q&A) during the solicitation period for questions and answers, and other significant information, relevant to the SBIR 09.2 topic under which they are proposing.

If you have general questions about DoD SBIR program, please contact the DoD SBIR Help Desk at (866) 724-7457 or email weblink.