Information and Decision Recommender
Navy STTR FY2014A - Topic N14A-T024
ONR - Steve Sullivan - email@example.com
Opens: March 5, 2014 - Closes: April 9, 2014 6:00am EST
N14A-T024 TITLE: Information and Decision Recommender
TECHNOLOGY AREAS: Information Systems, Human Systems
ACQUISITION PROGRAM: PMW150, SPAWAR, MC3 MARCORSYSCOM, PMMI MARCORSYSCOM, PMW 120
OBJECTIVE: The objective of this topic is to develop an effective and continuous operations and intelligence picture that supports decision making. To this end, an information and decision recommender system will be matured.
DESCRIPTION: Due to the large quantities of data now available to the warfighter it has become more difficult for the warfighter to find data that is most relevant to a course of action (COA) decision. Additionally, the development of COA recommendation tools has lagged the development of predictive tools related to enemy actions even though the two classes of tools may be able to leverage a common technology base. The objective of this topic is to address these two capability gaps by developing a recommender system for information delivery and COAs. For the purposes of this topic, proposers may assume that detailed COA can be represented by a simpler taxonomy consisting of "attack", "hold", "reinforce" and "retreat".
The challenge can be broken down into four parts. The first challenge lies in the development of an ability to map human entered COA to a machine understandable taxonomy (e.g. attack, hold, reinforce, and retreat). The second challenge is to map COA decision models to possible relevant features. Work related to feature selection that leverages modeling techniques for feature reduction can be leveraged . The third challenge is development of an information recommender system to personalize information delivery to specific warfighters making specific COA decisions based on past behavior. Such systems have been developed by the social networking community to optimize information delivery. Within those efforts, advances have been made in understanding the strengths and weaknesses of specific algorithms. Based on that understanding, recommender systems based on a fusion of recommendations have now been demonstrated . The fourth challenge is development of a system to incorporate a predictive capability that can suggest COA recommendations to an overtaxed commander based on the content of the information set identified as most relevant to a particular decision. The challenge of mapping observable or latent features to a COA recommendation accurately may require a layered modeling approach . The COA recommender must update recommendations whenever new and relevant data is received.
During Phase I, performers may select unclassified use cases from open source that have both rich reporting and a correct COA that became apparent after a period of time. Examples of this include political scandals (e.g. San Diego’s mayor’s eventual decision to retreat) or well documented military battles (e.g. Gettysburg).
The key technical challenges inherent to the topic include the development of 1) machine understanding of human entered COA, 2) automated feature selection and reduction, 3) algorithms that can tailor information selection and delivery, and 4) learning based decision recommendation tools.
PHASE I: Develop techniques to implement some or all of the component pieces of the described system; identify key technical risks associated with the development of a prototype; implement a design strategy to measure algorithm performance over time. Technical approach should address the challenges of 1) machine understanding of human entered courses of action, 2) automated feature selection and reduction, 3) algorithms that can tailor information selection and delivery, and 4) learning based decision recommendation tools. The Phase I effort should also identify a specific application and use case for a customer (military and commercial) and outline a plan for going forward with research. The final Phase I brief should include a proof of concept demonstration and show plans for a Phase II.
PHASE II: Produce an information and decision recommender prototype system that is capable of supporting a diverse set of COA. The prototype system should allow human entered COA to be entered for a specific mission and return COA recommendations. The recommended COA needs to be dynamically adjusted as new data becomes available during mission execution. The prototype should present pedigree information on recommendations with traceability back to key data features. During the Phase II effort, the transition path should be strengthened by focusing on data and COA trade spaces of interest.
PHASE III: Produce an application or set of applications that are capable being generalized to all Naval mission areas. The Phase III product(s) should be capable of running on program of record command and control systems while supporting data discovery on program of record intelligence systems within the Department of the Navy. The developed system must have relevance to tactical amphibious warfare and anti-access/area denial mission areas. During this phase the performer should concentrate on operational relevance and transition.
PRIVATE SECTOR COMMERCIAL POTENTIAL/DUAL-USE APPLICATIONS: The information and decision recommender system to be developed could be used by the private sector to develop decision support products related to employment and business decisions in a very similar manner.
2. Seok Jong Yu, "The Dynamic Competitive Recommendation Algorithm in Social Network Services", Information Sciences, 187 (2012) 1-14.
3. Yoshue Bengio, Learning Deep Architectures for AI, Foundations and trends in Machine Learning, Vol. 2, No. 1 (2009) 1-127.
4. Accumulo, http://accumulo.apache.org/.
KEYWORDS: Decision support, feature selection, predictive analysis, courses of action, fusion, recommender