System and method for quantifying and detecting non-normative behavior

A method for quantifying and detecting non-normative behavior, comprising the steps of:

• (a) providing a service system to collect, profile, and assess subject object behavior, said system identifying objects as discrete entities that involve interactions from different subjects and are extractable from on-line transaction events;
• (b) representing the similarity between two objects by coincidence in time by subject of the corresponding transaction events of those objects, aggregated across all subjects;
• (c) providing objects represented in vector form;
• (d) aggregating transaction events by subject to determine the subject's affinity to each object that the subject has interacted with;
• (e) providing subject vectors derived solely from said subject's transactions with said objects;
• (f) filtering by matching a vector of a subject against subject vectors of persons of interest;
• (g) determining a threshold to assess whether said matching falls within a pre-set detection range;
• and
• (h) triggering further investigation or testing against a population of normal subjects when said matching falls within said pre-set detection range,
• whereby said method quantifies and detects non-normative behavior by applying a suspicion threshold, and
• further whereby the subject vector and the object vector each have a respective number of dimensions;
• and
• further whereby the predicted similarity of one object to another object is calculated by matching their object vectors;
• and
• further whereby object vectors are generated by producing object vectors having respective initial dimensions, to determine predicted similarity values based on the initial object vectors, and to calculate a cost function that measures the difference between the predicted similarity values and the said similarity between two objects by coincidence in time by subject;
• and
• further whereby the dimensions of the generated object vectors are iteratively increased, and the cost function is recalculated based on the differences between the predicted similarity values and actual similarity values, until the cost function reaches a predetermined value, and wherein the actual similarity values are based on said transaction events;
• and
• further whereby the subject vectors are generated from the said object vectors and said affinities derived from said transaction events.

A method for using a system for quantifying and detecting non-normative behavior, comprising the steps of:

• (a) collecting subject object behavior;
• (b) profiling subjects and objects by generating subject and object vectors;
• (c) matching test subject profiles to target population subject profiles;
• and
• (d) detecting abnormality according to comparison of test target match to a pre-set detection range,
• whereby behavior data is collected electronically in an environment that provides a consistent record of behavior, said environment comprising a website, video surveillance, card access, phone records, or purchase histories, and
• further whereby the subject vector and the object vector each have a respective number of dimensions;
• and
• further whereby the predicted similarity of one object to another object is calculated by matching their object vectors;
• and
• further whereby object vectors are generated by producing object vectors having respective initial dimensions, to determine predicted similarity values based on the initial object vectors, and to calculate a cost function that measures the difference between the predicted similarity values and the said similarity between two objects by coincidence in time by subject;
• and
• further whereby the dimensions of the generated object vectors are iteratively increased, and the cost function is recalculated based on the differences between the predicted similarity values and actual similarity values, until the cost function reaches a predetermined value, and wherein the actual similarity values are based on said transaction events;
• and
• further whereby the subject vectors are generated from the said object vectors and said affinities derived from said transaction events, and without parsing transaction contents, all accomplished in a non-invasive, mentor-less fashion and being scalable to large numbers of said subjects and said objects.