Method and system for reducing risk values discrepancies between categories

Patent metadata

Field	Value
Publication number	US-11410062-B2
Application number	US-201715846500-A
Country	US
Kind code	B2
Filing date	Dec 19, 2017
Priority date	Dec 19, 2017
Publication date	Aug 9, 2022
Grant date	Aug 9, 2022

Abstract

The present teaching generally relates to removing perturbations from predictive scoring. In one embodiment, data representing a plurality of events detected by a content provider may be received, the data indicating a time that a corresponding event occurred and whether the corresponding event was fraudulent. First category data may be generated by grouping each event into one of a number of categories, each category being associated with a range of times. A first measure of risk for each category may be determined, where the first measure of risk indicates a likelihood that a future event occurring at a future time is fraudulent. Second category data may be generated by processing the first category data and a second measure of risk for each category may be determined. Measure data representing the second measure of risk for each category and the range of times associated with that category may be stored.

First claim

We claim:

1. A method for removing perturbations from predictive scoring, the method being implemented on at least one computing device having at least one processor, memory, and communications circuitry, and the method comprising:
allocating at least a portion of a memory for a data structure including first category data associated with a plurality of events detected by a content provider, the data structure including, for each event, an identifier, temporal information, an indication of one of a number of categories to which the event belongs, and an indication whether the event was fraudulent, each category being associated with a range of times;
determining, based on information stored in the data structure, a first measure of a risk for each category, wherein the first measure of the risk indicates a likelihood that a future event occurring at a future time within a range of times associated with the category is fraudulent;
generating second category data for obtaining a second measure of risk for each category, wherein the second measure of risk mitigates a transition of the corresponding first measure of risk of the category and the first measure of risk for an adjacent category, and wherein the category corresponds to a first range of times and the adjacent category corresponds to a second range of times adjacent to the first range of times;
adjusting the first measure of risk based on the second category data;
determining, based on the adjusted first measure of risk, the second measure of the risk for each category, wherein the second measure indicates an adjusted likelihood; and
storing, in a measure risk database, the second measure of risk for each category and the range of times associated with that category for future use by one or more predictive models to determine a likelihood that a future event occurring at a future time is fraudulent.

2. The method of claim 1 , wherein generating the second category data by processing the first category data comprises:
applying a smoothing function to the first category data, wherein the smoothing function is operable to mitigate transitions between a first value of the first measure of risk for one category and a second value of the first measure of risk for an adjacent category for the transitions that exceed a preset transition threshold value.

3. The method of claim 2 , further comprising:
determining a transition value associated with each of the transitions for each category;
determining, for each category, whether the transition value exceeds the preset transition threshold value; and
obtaining, in response to a corresponding transition value associated with one of the categories being determined to exceed the present transition threshold value, the smoothing function.

4. The method of claim 1 , wherein one of the categories comprises the future time.

5. The method of claim 1 , further comprising:
receiving new data representing a plurality of new events detected by the content provider, the new data indicating a new time that a corresponding new event occurred and whether the corresponding new event was fraudulent, wherein the plurality of new events corresponds to a duration of times subsequent to a previous duration of times associated with the data previously received;
generating third category data comprising by grouping each of the plurality of new event into one of the number of categories;
determining a third measure of the risk for each category; and
determining whether a difference between the third measure of risk for each category and a corresponding second measure of risk for each category exceeds a predetermined measure of risk threshold value, indicating that the third category data is to be ignored.

6. The method of claim 5 , further comprising:
determining a fourth measure of risk for each category based on, for each category, the first measure of risk, the second measure of risk, and the third measure of risk; and
storing new measure data representing the fourth measure of risk for each category and the range of times associated with that category.

7. A system for removing perturbations from predictive scoring, the system comprising:
a data bin filling system implemented by a processor and configured to:
allocate at least a portion of a memory for a data structure including first category data associated with a plurality of events detected by a content provider, the data structure including, for each event, an identifier, temporal information, an indication of one of a number of categories to which the event belongs, and an indication whether the event was fraudulent, each category being associated with a range of times, and
generate second category data for obtaining a second measure of risk for each category, wherein the second measure of risk mitigates a transition of the corresponding first measure of risk of the category and the first measure of risk for an adjacent category, and wherein the category corresponds to a first range of times and the adjacent category corresponds to a second range of times adjacent to the first range of times;
a measure of risk determination system implemented by the processor and configured to:
determine, based on information stored in the data structure, a first measure of a risk for each category, wherein the first measure of the risk indicates a likelihood that a future event occurring at a future time within a range of times associated with the category is fraudulent,
adjust the first measure of risk based on the second category data, and
determine, based on the adjusted first measure of risk, the second measure of the risk for each category, wherein the second measure indicates an adjusted likelihood; and
a measure risk database implemented by the processor and configured to store the second measure of risk for each category and the range of times associated with that category for future use by one or more predictive models to determine a likelihood that a future event occurring at a future time is fraudulent.

8. The system of claim 7 , further comprising:
a data smoothing application system implemented by the processor and configured to apply a smoothing function to the first category data, wherein the smoothing function is operable to mitigate transitions between a first value of the first measure of risk for one category and a second value of the first measure of risk for an adjacent category for the transitions that exceed a preset transition threshold value.

9. The system of claim 8 , wherein the measure of risk determination system is implemented by the processor and further configured to:
determine a transition value associated with each of the transitions for each category;
determine, for each category, whether the transition value exceeds the preset transition threshold value;
obtain, in response to a corresponding transition value associated with one of the categories being determined to exceed the present transition threshold value, the smoothing function.

10. The system of claim 7 , wherein one of the categories comprises the future time.

11. The system of claim 7 , further comprising:
a user event detection system implemented by the processor and configured to receive new data representing a plurality of new events detected by the content provider, the new data indicating a new time that a corresponding new event occurred and whether the corresponding new event was fraudulent, wherein the plurality of new events corresponds to a duration of times subsequent to a previous duration of times associated with the data previously received,
wherein the data bin filling system is implemented by the processor and further configured to generat

Assignees

• Yahoo Holdings Inc
• Yahoo Ad Tech Llc

Inventors

• Wang Liang
• Qiu Angus Xianen
• Pan Shengjun

Classifications

• G06N7/01 Primary

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06Q30/0248 Primary

  Avoiding fraud · CPC title

• G06Q20/4016

  involving fraud or risk level assessment in transaction processing · CPC title

• G06N5/022

  Knowledge engineering; Knowledge acquisition · CPC title

• G06N20/00

  Machine learning · CPC title