SEMANTIC CONTENT SEARCHING

• US 20110314024A1
• Filed: 06/18/2010
• Published: 12/22/2011
• Est. Priority Date: 06/18/2010
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Chang, Eric I-Chao, Xu, Yan, Feied, Craig, Gillam, Michael Todd, Handler, Jonathan
• Granted Patent: US 8,380,719 B2
• Time in Patent Office: Days
• US Class Current: 707/740
• CPC Class Codes: G06F 16/3325 Reformulation based on results of preceding query

Abstract

One or more techniques and/or systems are disclosed that provide for document retrieval where a user can identify key attributes of potential target documents that are desirable (e.g., have a particular semantic content for the user). Further, relevant documents that comprise the desired semantic content can be retrieved. Additionally, the user can provide feedback on the retrieved documents, for example, based on key semantic concepts found in the documents, and the input can be used to update the classification. For example, this process can be iterated to improve the retrieval and precision of documents found through machine learning techniques.

Claims

1. A computer-based method for document searching by semantic content, comprising:

receiving an end user selection of a desired first portion of an initial document from a database comprising potential target documents, the initial document comprising metadata labels that describe attributes of components of the initial document, the selected first portion comprising components of the initial document that have desired semantic content;

running the initial document, comprising the selected first portion, through one or more trained classifiers, using a computer-based processor, to identify a first potential target document from the database comprising a second portion that has a same semantic content as the first portion; and

if the second portion does not have the same semantic content as the first portion;

receiving an end user selection of a third portion of the first potential target document that comprises the same semantic content as the first portion; and

running the first potential target document, comprising the selected third portion, through the one or more trained classifiers to identify a second potential target document from the database comprising a fourth portion that has the same semantic content as the third portion.

2. The method of claim 1, comprising populating the database with potential target documents by performing a keyword search on a collection of documents.

3. The method of claim 1, if the second portion has the same semantic content as the first portion, comprising:

receiving an end user indication that the second portion is a correct selection; and

running the one or more classifiers over the database to select a third potential target document.

4. The method of claim 1, if the second portion does not have the same semantic content as the first portion and the first potential target document does not comprise content that has the same semantic content of the first portion, comprising:

receiving an end user indication that the first potential target document does not comprise content that has the same semantic content of the first portion; and

running the one or more classifiers over the database to select a third potential target document.

5. The method of claim 1, comprising utilizing user input for potential target documents returned by the one or more classifiers to update the one or more classifiers.

6. The method of claim 1, comprising running a plurality of documents through the one or more classifiers until a desired threshold is reached.

7. The method of claim 6, comprising receiving end user input for respective documents, the input comprising one or more of:

an indication that a document retrieved from the database by the one or more classifiers comprises desired semantic content;

an indication that the document retrieved from the database by the one or more classifiers does not comprise desired semantic content; and

a selected portion of the document retrieved from the database by the one or more classifiers, where the selected portion comprises desired semantic content.

8. The method of claim 6, running the plurality of documents through the one or more classifiers until a desired threshold is reached comprising one of:

running the plurality of documents through the one or more classifiers until a desired document selection accuracy is reached for desired semantic content;

running the plurality of documents through the one or more classifiers until a desired number of correct documents are retrieved without an incorrect document being retrieved; and

running the plurality of documents through the one or more classifiers until a desired number of documents have been retrieved from the database.

9. The method of claim 1, comprising presenting a potential target document to the end user merely when document classification provides a result that is within a desired threshold.

10. The method of claim 9, the desired threshold comprising an uncertainty threshold for a classifier model.

11. The method of claim 1, comprising utilizing a plurality of trained classifiers concurrently to retrieve potential target documents from the database.

12. The method of claim 11, comprising:

identifying which of the plurality of classifiers has a desired accuracy rate for retrieving potential target documents for a desired semantic content; and

utilizing the identified classifier to retrieve documents from the database for the desired semantic content.

13. The method of claim 11, comprising one or more of:

using a second classifier to cross-validate a potential target document retrieved by a first classifier; and

identifying which combination of two or more classifiers of the plurality of classifiers has a desired accuracy rate for retrieving potential target documents for a desired semantic content, and utilizing the identified combination of classifiers to retrieve documents from the database for the desired semantic content.

14. A system for document searching by semantic content, comprising:

a memory component configured to store a database comprising a plurality of potential target documents;

a processor component operably coupled with the memory component and configured to execute instructions for one or more classifiers;

an end-user input receiving component configured to receive end-user input for a document, the input comprising one or more of;

end user selection of a desired portion of a first document from the database, the selected portion comprising document components of the initial document that have desired semantic content;

end user indication that a second document retrieved from the database comprises a same semantic content as the selected desired portion of the first document; and

end user indication that a second document retrieved from the database does not comprise a same semantic content as the selected desired portion of the first document;

one or more classifier components operably coupled with the processor component and memory component, and configured to identify a second document from the database comprising a target portion that has a same semantic content as the selected desired portion of the first document; and

a classification updating component operably coupled with the end-user input receiving component and configured to utilize the end-user input to update the one or more classifier components to identify desired semantic content.

15. The system of claim 14, comprising a database populating component configured to populate the database with potential target documents using a keyword search of a collection of documents.

16. The system of claim 14, comprising a document presentation component, configured to present a document from the database to the end user for identification of desired semantic content.

17. The system of claim 14, the database comprising one or more sample documents that comprise the desired semantic content, the one or more sample documents configured to train the one or more classifiers to identify potential target documents having a same semantic content as the desired semantic content.

18. The system of claim 14, the desired portion of a document comprising text, where document components comprise words.

19. The system of claim 14, comprising a database indexing component configured to label a document in the database that comprises desired semantic content.

20. A method for document searching by semantic content, comprising:

populating a database with potential target documents by performing a keyword search on a collection of documents

receiving an end user selection of a desired first portion of an initial document from the database comprising potential target documents, the initial document comprising metadata labels that describe attributes of components of the initial document, the selected first portion comprising components of the initial document that have desired semantic content;

running the initial document, comprising the selected first portion, through one or more trained classifiers, using a computer-based processor, to identify a first potential target document from the database comprising a second portion that has a same semantic content as the first portion;

if the second portion does not have the same semantic content as the first portion;

receiving an end user selection of a third portion of the first potential target document that comprises the same semantic content as the first portion; and

running the first potential target document, comprising the selected third portion, through the one or more trained classifiers to identify a second potential target document from the database comprising a fourth portion that has the same semantic content as the third portion;

if the second portion has the same semantic content as the first portion;

receiving an end user indication that the second portion is a correct selection; and

running the one or more classifiers over the database to select a third potential target document;

if the second portion does not have the same semantic content as the first portion and the first potential target document does not comprise content that has the same semantic content of the first portion;

receiving an end user indication that the first potential target document does not comprise content that has the same semantic content of the first portion; and

running the one or more classifiers over the database to select a third potential target document;

utilizing user input for potential target documents returned by the one or more classifiers to update the one or more classifiers; and

running the plurality of documents through the one or more classifiers until a desired threshold is reached comprising one of;

running the plurality of documents through the one or more classifiers until a desired document selection accuracy is reached for desired semantic content;

running the plurality of documents through the one or more classifiers until a desired number of correct documents are retrieved without an incorrect document being retrieved; and

running the plurality of documents through the one or more classifiers until a desired number of documents have been retrieved from the database;

identifying which of the one or more classifiers has a desired accuracy rate for retrieving potential target documents for a desired semantic content; and

utilizing the identified classifier to retrieve documents from the database for the desired semantic content.