Surety is performing system maintenance this weekend. Electronic date stamps on new Prior Art Database disclosures may be delayed.
Browse Prior Art Database

Learning Model for Chart Visualization

IP.com Disclosure Number: IPCOM000234699D
Publication Date: 2014-Jan-29
Document File: 2 page(s) / 28K

Publishing Venue

The IP.com Prior Art Database


A method for creating a learning model to determine the best chart type to utilize for presenting data 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

Learning Model for Chart Visualization

Disclosed is a method for creating a learning model to determine a chart type to utilize for presenting data.

Business users as well as other users including data analysts would like to see data in a form that presents the data in a meaningful way, that is easy to understand and read in order to get insight into the meaning of the information. Previously, the business user or a data analyst, was required to specify the ideal visualization for a type of data ahead of time (just-in-time or manually pre-set).

The disclosed system and method monitors a history and applies previous selection to learn to choose the format for presenting the data (ex: pie/bar/line chart, graphs) for content based on the context, summary, statements and questions about the data from a user.

This learning model can then be used when presenting data based on questions from a user about the data, or for summarizing the data.

In one implementation during the ingestion of corpora the set of terms and keywords that are in text surrounding the area of presented data is denoted. The set of terms and keywords are stored in a set of n-grams with relations to the chart, the key words and phrases mined from the surrounding text. These n-grams are analyzed against data-statement-terms and added to the list of data-statement-terms based on their document frequency and association with the chart type. Based on these set of terms, weights are assigned to a learning model and chart type. A function is utilized for predicting the best chart based on the set of features defined and classified during ingestion. The documents can be user annotated if a chart type is not stated in text, or easily distinguishable during text or OCR analysis. The set of terms and chart associations are tied to domains and are then constructed in a learning model to yield the chart type per domain or general domain.

The system distinguishes and develops short phrases that fall into categories by utilizing example terms, for example, "how many", {"noun by noun", "term per term}. The terms are analyzed and associated with a chart during ingestion. It is expected that an accurate model for association can be constructed that would associate statements and keywords with a chart type based on the domain type or general domain.

The disclosed method builds up a categorization schema around the different chart types:

For example:


Chart Bar


Then the category schema and elements are given weights to the actual values in the mapping.

An expert...