System for generating scores related to interactions with a revenue generator

Abstract

A system is disclosed for generating scores related to interactions with a revenue generator. A historical dataset corresponding to a historical behavior of a set of revenue generators may be identified. The historical dataset may be processed to identify a feature vector. A classifier model may be generated from the historical dataset and the feature vector. Current revenue generator data representing a current revenue generator may be collected. The current revenue generator data may be processed to generate a current revenue generator feature vector. A score may be generated by applying the classifier model to the current revenue generator data feature vector.

We claim: 
     
         1 . A system for generating scores for a revenue generator, comprising:
 a memory to store a classifier model, a historical revenue generator dataset related to an advertising system, a feature vector, current revenue generator data from a current revenue generator, and a current revenue generator data feature vector;   wherein the feature vector comprises at least two variables from the historical revenue generator dataset;   wherein the current revenue generator data feature vector comprises at least two variables from the current revenue generator data;   an interface coupled with the memory and configured to collect the current revenue generator data from the current revenue generator; and   a processor coupled with the memory and the interface, that is configured to:
 process the historical revenue generator dataset to generate the feature vector, 
 generate the classifier model from the historical dataset and the feature vector, 
 process the current revenue generator data to generate the current revenue generator data feature vector, and 
 generate a score representing a probability of the current revenue generator committing fraud by applying the classifier model to the current revenue generator data feature vector. 
   
     
     
         2 . The system of  claim 1  wherein the classifier model is generated by using a machine learning algorithm. 
     
     
         3 . The system of  claim 1  wherein the processor modifies the current revenue generator data to include a classification value indicative of whether the current revenue generator committed fraud, adds the current revenue generator data to the historical revenue generator dataset, reprocesses the historical revenue generator dataset to generate an updated feature vector comprising an updated at least two variables, and re-generates the classifier model from the historical revenue generator dataset and the updated feature vector. 
     
     
         4 . The system of  claim 1  wherein the processor receives an indication of an amount of functionality of the advertising system that should be provided to the current revenue generator, and provides to the current revenue generator the amount of functionality of the advertising system indicated. 
     
     
         5 . The system of  claim 4  wherein the processor updates a status of the revenue generator based on the score, wherein the status is indicative of the amount of functionality of the advertising system provided to the current revenue generator. 
     
     
         6 . The system of  claim 1  wherein the processor receives an update to the current revenue generator data and modifies the current revenue generator data based on the received update. 
     
     
         7 . The system of  claim 6  wherein the update comprises an update to a spend limit of the current revenue generator data, wherein the spend limit indicates an amount of charges the current revenue generator can accrue in the advertising system over a period of time, updates the spend limit of the current revenue generator data based on the received update, and limits the amount of charges the current revenue generator may accrue over the period of time. 
     
     
         8 . The system of  claim 1  wherein the processor is further configured to:
 receive an indication that that the score for the current revenue generator does not accurately reflect the behavior of the current revenue generator related to the advertising system; 
 receive an update to the historical revenue generator dataset, wherein the update comprises data describing the current revenue generator and an associated classification indicating whether the current revenue generator committed fraud; 
 add the update to the historical dataset; 
 reprocess the historical dataset to generate an updated feature vector comprising an updated at least two variables; and 
 regenerate the classifier model from the reprocessed historical dataset and the updated feature vector, wherein the set of inputs of the regenerated classifier model comprises the at least two variables of the updated feature vector. 
 
     
     
         9 . The system of  claim 1  wherein the feature vector comprises at least one of a spend history, a spend to replenish ratio, an average amount of payment, a number of times a credit card is charged in a month, a total sum of charges accrued in a month, an average adjustment amount, a total number of adjustments, a client tier value, a security status value, a risk status value, a client age value, a client search term max-spend-score, a client search term risk score, a client search term risk-spend score, a client max spend over daily budget score, a client run rate over daily budget score, a client risk-run rate score, a client run-rate rate of change score, and a client card score. 
     
     
         10 . A non-transitory computer readable medium having stored therein data representing instructions executable by a programmed processor for generating scores for a revenue generator, storage medium comprising instructions operative for:
 identifying a historical dataset corresponding to a historical behavior of a set of revenue generators in an advertising system;   processing the historical dataset to generate a feature vector comprising at least two variables;   generating a classifier model from the historical dataset and the feature vector, wherein a set of inputs of the generated classifier model comprises at least two variables of the feature vector;   collecting current revenue generator data representing a behavior of a current revenue generator in the advertising system;   processing the current revenue generator data to generate a current revenue generator data feature vector, wherein the current revenue generator data feature vector comprises at least two variables of the current revenue generator data; and   generating a score for the current revenue generator by applying the classifier model to the current revenue generator data feature vector, wherein the score represents a probability that the current revenue generator will commit fraud.   
     
     
         11 . The storage medium of  claim 10  where in the historical behavior of the set of revenue generators is identified as fraudulent behavior or not fraudulent behavior. 
     
     
         12 . The storage medium of  claim 10  wherein the revenue generator comprises an online advertiser. 
     
     
         13 . The storage medium of  claim 10  further comprising receiving an update to the current revenue generator data and modifying the current revenue generator data based on the received update. 
     
     
         14 . A method of scoring a revenue generator, comprising:
 collecting revenue generator data representing a revenue generator;   processing the revenue generator data to generate a revenue generator feature vector comprising a combination of at least two values of the revenue generator data; and   generating, by at least one processor, a score of the revenue generator data indicating a probability of the revenue generator being fraudulent by applying the revenue generator feature vector to a classifier model.   
     
     
         15 . The method of  claim 14  further comprising:
 modifying the revenue generator data to include a classification value, wherein the classification value is indicative of whether the revenue generator committed fraud; 
 adding the revenue generator data to a historical revenue generator dataset; 
 processing the historical revenue generator dataset to generate an updated feature vector comprising an updated at least two variables, wherein the updated at least two variables for data of a second revenue generator is indicative of a probability that the second revenue generator is fraudulent; and 
 re-generating the classifier model from the historical revenue generator data and the updated feature vector. 
 
     
     
         16 . The method of  claim 16  further comprising generating the classifier model from a historical revenue generator data and a feature vector comprising at least two variables. 
     
     
         17 . The method of  claim 17  wherein generating the score further comprises:
 generating a scoring metric; and 
 applying the scoring metric to the score generated by applying the feature vector of the revenue generator to the classifier model. 
 
     
     
         18 . The method of  claim 17  wherein the scoring metric comprises a multiplier. 
     
     
         19 . The method of  claim 15  further comprising receiving an update to the revenue generator data and modifying the revenue generator data based on the received update. 
     
     
         20 . The method of  claim 14  wherein the revenue generator comprises an online advertiser.