Browse Prior Art Database

Optimal simulation of real world transactions from given (decision tree based) business patterns or small sample of domain transactions

IP.com Disclosure Number: IPCOM000211466D
Original Publication Date: 2011-Oct-05
Included in the Prior Art Database: 2011-Oct-05
Document File: 6 page(s) / 244K

Publishing Venue

Microsoft

Related People

Partha Pratim Ghosh: INVENTOR [+5]

Abstract

Recommender System is a key component in analysing consumers’ transactions to discover insightful latent business patterns (base patterns) aka rules. However, validation of such a system requires humongous real world data representing typical behavior of consumers. This is either too expensive or too less to be of use. The ability to create such transactions based on patterns created by experts in a particular domain or from a small domain sample, hence, cannot be overemphasized. A key challenge while generating such transactions from base patterns provided by domain experts, is to be able to inject controlled ‘noise’ (i.e. transactions that do not follow the business patterns) without upsetting the statistics of transactions respecting the base patterns. Injection of optimal noise is key to understanding the discerning power of a recommender system. We have invented a generic technique to simulate transactions optimally and accurately given a set of patterns for a domain. This is extremely useful for validation of a recommendation system designed to discover interesting patterns as access to real data is costly or unavailable. The generated data must be obfuscated with noise (noisy rules). We constructed a proof of its optimality. In cases, where patterns are not known in advance but small real world sample data is available, we have provided an alternative to discover the base patterns from the same. This approach is potentially applicable to any domain where patterns are either known in advance or can be created from a small sample of real world data.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 29% of the total text.

Document Author (alias)

Parthag

Defensive Publication Title 

Optimal simulation of real world transactions from given (decision tree based) business patterns or small sample of domain transactions

Name(s) of All Contributors

1.    Partha Pratim Ghosh  (Alias: parthag@microsoft.com)

2.    Shradha Chaudhary (Alias: shradhac@microsoft.com )

3.    Nagendra Kumar (Alias: nakuma@microsoft.com)

4.    Hrushikesh Bokil (Alias: hrushib@microsoft.com)

5.    Vinay Singh (Alias: vinaysin@microsoft.com)

Summary of the Defensive Publication/Abstract

Recommender System is a key component in analysing consumers’ transactions to discover insightful latent business patterns (base patterns) aka rules. However, validation of such a system requires humongous real world data representing typical behavior of consumers. This is either too expensive or too less to be of use. The ability to create such transactions based on patterns created by experts in a particular domain or from a small domain sample, hence, cannot be overemphasized. 

A key challenge  while generating such transactions from base patterns provided by domain experts, is to be able to inject controlled  ‘noise’ (i.e. transactions that do not follow the business patterns) without upsetting the statistics of transactions respecting the base  patterns. Injection of optimal noise is key to understanding the discerning power of a recommender system.  

We have invented a generic technique to simulate transactions optimally and accurately given a set of patterns for a domain. This is extremely useful for validation of a recommendation system designed to discover interesting patterns as access to real data is costly or unavailable. The generated data must be obfuscated with noise (noisy rules). We constructed a proof of its optimality. In cases, where patterns are not known in advance but small real world sample data is available, we have provided an alternative to discover the base  patterns from the same. This approach is potentially applicable to any domain where patterns are either known in advance or can be created from a small sample of real world data.

Description:  Include architectural diagrams and system level data flow diagrams if: 1) they have already been prepared or 2) they are needed to enable another developer to implement your defensive publication. Target 1-2 pages, and not more than 5 pages.  

Theory: Complementary Rules Generation

Given support and confidence (See attached whitepaper) of a set of business patterns aka rules (aka signal), we derive the minimum number of complementary (aka noise)  rules necessary to obfuscate generated transactions for challenging a recommendation system. Our algorithm clearly articulates how to assign (support, confidence) for each of the complementary rules including their forms (Antecedent →Consequent). As far as we know, there was no such work has been done towards determining the structure of the complementary rules in an optimal manner. Please find...