True Colors: Persona/Human Network Similarity Quantification Software

In response to Navy SBIR topic N092-149, “Similarity Measures for Persona/Human Networks”, Michigan Aerospace Corporation (MAC) proposes the True Colors system (TC): an application that utilizes a unique means of assessing, quantifying and understanding the relationship between N-dimensional vectors which represent signatures capturing the behavior and activities of individuals or groups. Unlike popular distance and similarity metrics, the TC technique for assessing similarity does not require normalization, works with mixed numeric and categorical fields, gracefully handles missing entries, and can function appropriately in the case of imbalanced or sparsely populated records. TC is based on several years of research conducted by MAC staff in conjunction with government and industry partners in applying Ensembles of Decision Trees (EDTs) to challenging problems in Data Mining and Pattern Recognition including clustering. Advantages to this approach, include confidence estimations and relative feature importance assessment. MAC will use anonymized human interactions harvested from the well-known Slashdot forums as a testbed to demonstrate TC similarity principles. Slashdot forums supply a wealth of information about authors, users, topics, and cliques of like-minded members. The ability to assess similarity between items of varied composition has key applications in business, biology, social network analysis and many other fields.