Text generator with an automated decision tree for creating text based on changing input data

• US 7,376,552 B2
• Filed: 08/12/2003
• Issued: 05/20/2008
• Est. Priority Date: 08/12/2003
• Status: Active Grant

• Current Assignee: Markit on Demand, Inc. (S&P Global Inc.)
• Original Assignee: Wall Street On Demand
• Inventors: Hander, William B., De La Lama, Ariel, Levine, Fredrick E., Leslie, John W., Tanner, James
• Primary Examiner(s): Vo; Huyen X.
• Application Number:
  US10/639,274
• Publication Number:
  US 20050039107A1
• Time in Patent Office: 1,743 Days
• Field of Search: 704 1- 10, 704/251, 704/252, 704/257, 704/270, 715/501.1
• US Class Current: 704/9
• CPC Class Codes: G06F 40/56 Natural language generation

Abstract

A method for generating text using a computer system. The method includes selecting a model for generating a text that is based on a decision tree with nodes having one buckets or elements that include a conditional element and a sentence element or a pointer to a child node. The method includes processing a set of raw data selected for the particular model and then determining values of model variables. Evaluation of the model is performed to identify a set of the sentence elements to include in the text. The evaluating includes processing the conditional element of the buckets and when the conditional element is satisfied including the bucket'"'"'s sentence element in the set of identified sentence elements. The method continues with combining the identified sentence elements to generate the text. The combining is completed based on concatenation rules provided in or linked to each node.

Claims

1. A computer-based method for generating text, comprising:

selecting a model of a text, the model being a decision tree with a node having one or more buckets including a conditional element and a sentence element;

processing a set of raw data selected based on the selected model to determine a model variable;

evaluating the selected text model to identify a set of the sentence elements to include in the generated text, wherein the evaluating includes processing the conditional element of a bucket and when the conditional element is satisfied including the sentence element for the bucket in the identified set of the sentence elements; and

combining the sentence elements in the identified set to form the generated text;

wherein the processing of one of the conditional elements includes retrieving the model variable determined from the processed set of raw data and comparing a value of the model variable with a predefined conditional value for the bucket and wherein each of the nodes of the selected text model include a set of concatenation rules for the sentence elements of the particular node and wherein the combining of the sentence elements includes concatenating the sentence elements for each of the nodes based on the sets of concatenation rules, the concatenation rules including definitions of connectors to be inserted between the sentence elements.

2. The method of claim 1, wherein the selected text model includes a plurality of the nodes each having one or more buckets pointing to another of the nodes or including one of the sentence elements.

3. The method of claim 1, further including receiving a user input comprising a unique key and retrieving the set of raw data from a data source based on the unique key.

4. The method of claim 3, wherein the user input further includes a model selection and the selecting includes processing the model selection to identify the selected model from a set of available text models.

5. The method of claim 1, further including transferring the formed text to a user node for display to a user.

6. The method of claim 1, wherein the set of raw data includes financial analysis information, the model variable is a financial analysis variable, and the sentence elements include text related to financial analysis.

7. A text generation system, comprising:

means for storing a set of text models, each of the text models being decision tree based and including a parent node and a plurality of child nodes each having one or more bucket elements, wherein at least some of the bucket elements include conditional elements and a sentence element;

means for storing a model database including retrieved raw data and model variables for a selected one of the text models;

a generator engine configured for identifying the selected one of the text models, for processing the retrieved raw data to determine the model variables, and for generating a text by processing the conditional elements to identify a set of the sentence elements and combining the identified sentence elements, wherein at least some of the buckets include multiple text strings and one of the text strings is randomly selected for inclusion in the text by the generator engine.

8. The system of claim 7, wherein the generator engine is further configured to receive a user input including a unique key and to select the retrieved raw data based on the unique key.

9. The system of claim 8, wherein the user input includes a model selection and the generator engine uses the model selection in performing the identifying of the selected one of the text models.

10. The system of claim 7, wherein at least some of the nodes include a set of concatenation rules and wherein the generator engine uses the concatenation rules during the combining of the identified sentence elements.

11. The system of claim 10, wherein the concatenation rules define connectors to be inserted by the generator engine between combined ones of the sentence element in the generated text.

12. A computer-based method for generating text, comprising:

receiving user input requesting that text be built, wherein the user input comprises a key and a model selection;

selecting a text model based on the model selection, the text model being a decision tree with nodes comprising buckets having a conditional element and a corresponding sentence element;

retrieve a set of raw data from a data source based on the key and on the text model;

processing the set of raw data to determine model variables defined by the text model;

evaluating the selected text model to identify a set of the sentence elements to include in the text, wherein the evaluating includes processing the conditional elements of at least some of the buckets using the determined model variables and when the conditional elements are satisfied including the corresponding sentence elements for the buckets in the identified set of the sentence elements; and

combining the sentence elements in the identified set to form the text;

wherein at least some of the buckets include multiple text strings and one of the text strings is randomly selected for inclusion in the text.

13. The method of claim 12, wherein each of the nodes of the text model includes a set of concatenation rules for the sentence elements of the particular node and wherein the combining of the sentence elements includes concatenating the sentence elements for each of the nodes separately based on the sets of concatenation rules.

14. The computer readable medium method of claim 13, wherein the concatenation rules include definitions of connectors to be inserted between the sentence elements and wherein at least one of the sets of the concatenation rules differs from at least another one of the sets of the concatenation rules, whereby combination of the sentence elements at the nodes is performed uniquely at the corresponding nodes.

15. The computer readable medium method of claim 12, wherein the key is a securities symbol, the set of raw data includes financial analysis information, the model variable is a financial analysis variable, and the sentence elements include text related to financial analysis.

16. The method of claim 12, wherein the one of the text strings is selected randomly and without repetition within the text.

17. The method of claim 12 wherein at least some of the sentence elements comprise one or more of the determined model variables, whereby the determined model variables are presented in the text.