InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Data Characteristic Reporting and Modeling Tuning Recommendation

IP.com Disclosure Number: IPCOM000238491D
Publication Date: 2014-Aug-28
Document File: 2 page(s) / 117K

Publishing Venue

The IP.com Prior Art Database


A method for providing data characteristic reporting and modeling tuning recommendations is disclosed.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 52% of the total text.

Page 01 of 2

Data Characteristic Reporting and Modeling Tuning Recommendation

Disclosed is a method for providing data characteristic reporting and modeling tuning recommendations.


        ® Retail Solution provides retailers with optimal pricing recommendations, which is built on a statistical model. The Modeling Service Statisticians are the users of this model, who run the model, validate the model, and release the model. The model is controlled by a list of settings, different run values of which will potentially generate different model fits, good or bad, based on characteristics of the data. While there's not an automatic detecting system for the best values of those modeling settings, statisticians have to wait for the completion of the model to examine the results and decide the modeling settings that need to be tuned. Yet the implementation of such tuning needs to be done via another round of remodeling effort. For example, a modeling option called "USE_CLUSTERS" (store clustering), is an option to aggregate store level data to cluster level data, such that the model could generate more reliable estimation off of richer data. This method can be leveraged to deal with data sparsity. However, there's lack of a way to automatically detect the data sparsity before a modeling job is run. Therefore, Statisticians have to go through several iterations and then decide whether clustering is necessary for a model.

Utilizing the disclosed method, the system can detect the sparsity issue based on the data characteristics beforehand and suggest using store clustering for a particular category, which could save a lot of statistician efforts as well as modeling iterations.

The following steps are performed:
Recycle the information as useful input for tuning modeling option. Take modeling setting "USE_CLUSTERS" for example, DB2* queries may be used






The features of this invention include:

Display of informative metrics and recommended tuning metrics from application allows different users to access information easily and provides with users better visibility to data characteristics of categories.

Automatic tuning of modeling settings on the fly.

Ability for the users to select multiple categories to run metrics before kicking off the model saves time and efforts.

Ability for the users to schedule the job enhances flexibility and manageability.

Data Characteristics includes two parts

Informative: Reporting metrics


Missing Weeks

DG Price Ratio
DG Size Ratio
Customer Promotion Information

Assortment Index

Model UI Parameters: Tuning recommendation NATL_PROMO (national promotion modeling option)

USE_CLUSTERS (use store clustering modeling option)

AdjustDgSalesByChunk (adjust demand group volume for large-assortment categories)

Figure 3 shows the contrast of the workflow before and after the implementation:

Before the implementation, as depicted in Figure 1, users (Statistician) need to...