SYSTEM AND METHODS FOR AUTOMATICALLY DETECTING DECEPTIVE CONTENT

• US 20140304814A1
• Filed: 10/18/2012
• Published: 10/09/2014
• Est. Priority Date: 10/19/2011
• Status: Active Grant

• Current Assignee: Cornell University
• Original Assignee: Cornell University
• Inventors: Cardie, Claire, Ott, Myle, Chol, Yejin, Hancock, Jeffrey
• Granted Patent: US 10,642,975 B2
• Time in Patent Office: Days
• US Class Current: 726/22
• CPC Class Codes: G06F 21/552 involving long-term monitoring or reporting
• CPC Class Codes: G06F 2221/034 Test or assess a computer or a system
• CPC Class Codes: G06F 40/253 Grammatical analysis; Style critique

Abstract

Systems and methods for detecting deceptive opinion spam. Certain embodiments include a classifier with improved accuracy for detecting deceptive opinion entries. A feature analysis of learned models reveals a relationship between deceptive opinions and imaginative writing. By modeling deception in a generative framework, the prevalence of deception in two popular online review communities may be determined. Deceptive opinion spam is a rapidly growing and widespread problem, especially in review communities with minimal posting requirements.

Claims

1. A system for automatically detecting deceptive textual information, comprising:

a processor;

a main memory in communication with the processor via a communication infrastructure and storing instructions that, when executed by the processor, cause the processor to;

collect opinion information to create an initial dataset;

analyze the initial dataset;

generate a model for detection of deceptive opinion information;

obtain new opinion information not in the initial dataset to form a second dataset;

apply the model to the second dataset to detect deceptive textual information within the second dataset.

2. The system of claim 1, further comprising:

a display unit;

a user interface configured to be displayed on the display unit, said user interface configured to show opinion information and to permit a user to select which opinion information is included in the initial dataset or the second dataset.

3. The system of claim 2 wherein the user interface also permits the user to activate or deactivate system responses to detection of deceptive information, said responses including identifying deceptive opinion information with an indicator displayed in the user interface or eliminating deceptive opinion information from the second dataset.

4. The system of claim 1, further comprising a secondary memory configured to permit storage of at least opinion information.

5. The system of claim 1, wherein, when stored instructions cause the processor to analyze the initial dataset, the analysis includes utilizing an analysis approach selected from the group consisting of a genre identification approach, a psycholinguistic approach, and a text categorization approach.

6. The system of claim 1, wherein, when stored instructions cause the processor to analyze the initial dataset, the analysis includes employing two or more analysis approaches, wherein each approach is selected from the group of genre identification approach, a psycholinguistic approach, a text categorization approach, a combined genre identification-psycholinguistic approach, a combined genre identification-text categorization approach, a combined psycholinguistic-text categorization approach, and a combined genre identification-psycholinguistic-text categorization approach, and the stored instructions cause the processor to also:

determine the accuracy of each of the two or more analysis approaches; and

compare the accuracy of each of the two or more analysis approaches.

7. The system of claim 1, wherein the stored instructions, when executed by to processor, also cause the processor to code the opinion information in the initial dataset.

8. The system of claim 1, wherein the stored instructions, when executed by to processor, also cause the processor to calculate statistics regarding the prevalence of and characteristics of deceptive opinion information.

9. A method for classifying textual opinion information as truthful or deceptive, comprising the steps of:

communicating with a first source of opinion information via a communication interface connecting a user system to a network to collect opinion information forming an initial dataset, the communication interface being a wired communication interface or a wireless communication interface;

storing the opinion information in a main memory of the user system;

machine-analyzing the opinion information in the initial dataset;

automatically generating a model for detection of deceptive opinion information;

obtaining new opinion information not in the initial dataset to form a second dataset; and

applying the model to the second dataset to classify as truthful or deceptive the textual information within the second dataset.

10. A method of claim 9, wherein said analyzing step includes utilizing an approach selected from the group consisting of a genre identification approach, a psycholinguistic approach, and a text categorization approach.

11. The method of claim 9, wherein said analyzing step includes using a combined approach that combines two or more approaches selected from the group of a genre identification approach, a psycholinguistic approach, and a text categorization approach.

12. A method of claim 9, further comprising the step of providing a user interface configured to be viewed in a display unit, the user interface showing opinion information and presenting user selection options for opinion information collection, analysis, and response.

13. A method of claim 12, wherein the user selection option for response includes at least:

designating a first indicator to signify “

deceptive”

classification for opinion information;

assigning the “

deceptive”

indicator to each opinion information entry detected by the model as deceptive; and

displaying each deceptive opinion information entry with the associated “

deceptive”

indicator.

14. A method of claim 13, wherein the user selection option for response permits the user to designate a second indicator for a “

false positive”

deceptive opinion information entry and a third indicator for a “

false negative”

deceptive opinion information entry.

15. A method of claim 13, wherein the user selection option for response includes at least:

designating a fourth indicator to signify “

truthful”

classification for opinion information;

assigning the “

truthful”

indicator to each truthful opinion information entry detected by the model as truthful; and

displaying each opinion information entry with the associated “

truthful”

indicator.

16. A method of claim 13, wherein the user selection option for response includes, upon categorization of the opinion information entry as “

deceptive”

, automatically removing an opinion information entry from the second dataset.

17. A computer-readable storage medium storing one or more sequences of instructions which, when executed by one or more processors, causes the one or more processors to perform the following steps:

obtaining opinion information to form a dataset; and

applying a deceptive opinion detection model including at least one of a utilizing a genre identification approach, a psycholinguistic approach, and a text categorization approach to the dataset to detect deceptive textual information.

18. A computer-readable storage medium storing one or more sequences of instructions of claim 17, which, when executed by one or more processors, causes the one or more processors to perform the following additional steps:

providing a user interface configured to be displayed on a display unit, said user interface configured to show opinion information and to permit a user to choose an output from the group of adding an indicator to a “

deceptive”

review or causing a “

deceptive”

review to be hidden from view in the user interface.

19. A computer-readable storage medium storing one or more sequences of instructions which, when executed by one or more processors, causes the one or more processors to:

receive a user input including at least an opinion information entry;

apply a deceptive opinion detection model including at least one of a utilizing a genre identification approach, a psycholinguistic approach, and a text categorization approach to the dataset to detect deceptive textual information; and

providing a user interface configured to be displayed on a display unit, said user interface configured to show opinion information and to permit a user to choose an output from the group of adding an indicator to a “

deceptive”

review or causing a “

deceptive”

review to be hidden from view in the user interface.

20. A computer-readable storage medium of claim 19, wherein the instructions are configured for integration in a review-based website.