Cloud-based plagiarism detection system performing predicting based on classified feature vectors

• US 9,514,417 B2
• Filed: 12/30/2013
• Issued: 12/06/2016
• Est. Priority Date: 12/30/2013
• Status: Expired due to Fees

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Kumar, Sanjiv, Kernighan, Brian
• Primary Examiner(s): Vincent, David
• Application Number:
  US14/143,710
• Publication Number:
  US 20150186787A1
• Time in Patent Office: 1,072 Days
• Field of Search: 706/12, 706/45
• US Class Current: 1/1
• CPC Class Codes: G06F 16/93 Document management systems
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06Q 10/107 Computer-aided management of electronic mailing [e-mailing]

Abstract

Plagiarism may be detected, as disclosed herein, utilizing a database that stores documents for one or more courses. The database may restrict sharing of content between documents. A feature extraction module may receive edits and timestamp the edits to the document. A writing pattern for a particular user or group of users may be discerned from the temporal data and the documents for the particular user or group of users. A feature vector may be generated that represents the writing pattern. A machine learning technique may be applied to the feature vector to determine whether or not a document is plagiarized.

Claims

1. A computer-implemented method comprising:

generating training data for training a predictive model using a machine learning technique to estimate a probability that a given document is plagiarized or is not plagiarized, the training data including, for each of a plurality of training documents, a feature vector that includes (i) data referencing a content of an edit to the training document, (ii) data referencing a type of the edit to the training document, (iii) data referencing a time associated with the edit to the training document, and (iv) a label indicating whether the training document is or is not plagiarized;

training the predictive model using the training data;

after training the predictive model, identifying a particular document stored in a database;

receiving data referencing (i) a content of an edit to the particular document stored in the database, and (ii) a time associated with the edit to the particular document;

generating a feature vector based at least on the data referencing (i) the content of the edit to the particular document stored in the database, and (ii) the time associated with the edit to the particular document; and

determining a probability that the particular document is plagiarized or is not plagiarized based on classifying the feature vector by the predictive model that is trained using the machine learning technique.

2. The method of claim 1, wherein the type of the edit comprises an insertion or a deletion.

3. The method of claim 1, wherein the type of edit comprises a move.

4. The method of claim 1, wherein the type of edit comprises a replacement.

5. The method of claim 1, comprising automatically generating the plurality of training documents.

6. The method of claim 1, comprising, for each of the plurality of training documents, pre-processing the content of the edit to insert variable-invariant features.

7. The method of claim 6, wherein pre-processing the content comprises substituting a particular term included in the content with a synonym of the particular term.

8. A system comprising:

a processor configured to execute computer program instructions; and

a computer storage medium encoded with the computer program instructions that, when executed by the processor, cause the system to perform operations comprising;

generating training data for training a predictive model using a machine learning technique to estimate a probability that a given document is plagiarized or is not plagiarized, the training data including, for each of a plurality of training documents, a feature vector that includes (i) data referencing a content of an edit to the training document, (ii) data referencing a type of the edit to the training document, (iii) data referencing a time associated with the edit to the training document, and (iv) a label indicating whether the training document is or is not plagiarized;

training the predictive model using the training data;

after training the predictive model, identifying a particular document stored in a database;

receiving data referencing (i) a content of an edit to the particular document stored in the database, and (ii) a time associated with the edit to the particular document;

generating a feature vector based at least on the data referencing (i) the content of the edit to the particular document stored in the database, and (ii) the time associated with the edit to the particular document; and

determining a probability that the particular document is plagiarized or is not plagiarized based on classifying the feature vector by the predictive model that is trained using the machine learning technique.

9. The system of claim 8, wherein the type of the edit comprises an insertion or a deletion.

10. The system of claim 8, wherein the type of edit comprises a move.

11. The system of claim 8, wherein the type of edit comprises a replacement.

12. The system of claim 8, wherein the operations comprise automatically generating the plurality of training documents.

13. The system of claim 8, wherein the operations comprise, for each of the plurality of training documents, pre-processing the content of the edit to insert variable-invariant features.

14. The system of claim 13, wherein pre-processing the content comprises substituting a particular term included in the content with a synonym of the particular term.

15. A computer-readable storage device encoded with a computer program, the computer program comprising instructions that, when executed by one or more computers, cause the one or more computers to perform operations comprising:

generating training data for training a predictive model using a machine learning technique to estimate a probability that a given document is plagiarized or is not plagiarized, the training data including, for each of a plurality of training documents, a feature vector that includes (i) data referencing a content of an edit to the training document, (ii) data referencing a type of the edit to the training document, (iii) data referencing a time associated with the edit to the training document, and (iv) a label indicating whether the training document is or is not plagiarized;

training the predictive model using the training data;

after training the predictive model, identifying a particular document stored in a database;

receiving data referencing (i) a content of an edit to the particular document stored in the database, and (ii) a time associated with the edit to the particular document;

generating a feature vector based at least on the data referencing (i) the content of the edit to the particular document stored in the database, and (ii) the time associated with the edit to the particular document; and

determining a probability that the particular document is plagiarized or is not plagiarized based on classifying the feature vector by the predictive model that is trained using the machine learning technique.

16. The device of claim 15, wherein the type of the edit comprises an insertion or a deletion.

17. The device of claim 15, wherein the type of edit comprises a move.

18. The device of claim 15, wherein the type of edit comprises a replacement.

19. The device of claim 15, wherein the operations comprise automatically generating the plurality of training documents.

20. The device of claim 15, wherein the operations comprise, for each of the plurality of training documents, pre-processing the content of the edit to insert variable-invariant features.