PSYCHO-LINGUISTIC STATISTICAL DECEPTION DETECTION FROM TEXT CONTENT

• US 20130046531A1
• Filed: 01/06/2011
• Published: 02/21/2013
• Est. Priority Date: 01/07/2010
• Status: Active Grant

• Current Assignee: The Trustees of The Stevens Institute of Technology (Stevens Institute Of Technology)
• Original Assignee: The Trustees of The Stevens Institute of Technology (Stevens Institute Of Technology)
• Inventors: Chandramouli, Rajarathnam, Subbalakshmi, Koduvayur P., Chen, Xiaoling
• Granted Patent: US 9,116,877 B2
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G06F 40/216 using statistical methods
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

An apparatus and method for determining whether a text is deceptive may comprise analyzing a body of textual content known to be one of text containing true content and text containing deceptive content; identifying psycho-linguistic cues that are indicative of a text being deceptive; statistically analyzing, via a computing device, a given text based upon the psycho-linguistic cues to determine if the text is deceptive. The apparatus and method may further comprise weighting the psycho-linguistic cues and statistically analyzing based on the weighted psycho-linguistic cues. The statistically analyzing step may be performed using one of a cue matching analysis, a weighted cue matching analysis, a Markov chain analysis, and a sequential probability ratio testing binary hypothesis analysis. The psycho-linguistic cues may be separated into categories, including increasing trend cues and decreasing trend cues and analyzed according to presence in a category from within the categories.

Claims

1. A method for determining whether a text is deceptive, comprising:

analyzing a body of textual content known to be one of text containing true content and text containing deceptive content;

identifying psycho-linguistic cues that are indicative of a text being deceptive;

statistically analyzing, via a computing device, a given text based upon the psycho-linguistic cues to determine if the text is deceptive.

2. The method of claim 1 further comprising:

weighting the psycho-linguistic cues and statistically analyzing based on the weighted psycho-linguistic cues.

3. The method of claim 1, wherein the statistically analyzing step is performed using one of a cue matching analysis, a weighted cue matching analysis, a Markov chain analysis, and a sequential probability ratio testing binary hypothesis analysis.

4. The method of claim 1 wherein the psycho-linguistic cues are selected from a list consisting only of more of the number of (1) words, (2) first person plural words, (3) inclusive words, (4) affective language words, (5) optimism and energy words, (6) social process words and less of the number of (7) diversity words, (8) first person singular pronouns, (9) third person pronouns, (10) assent language words, (11) insight words, (12) tentative words and (13) exclusive words, (14) past tense verbs, (15) present tense verbs and (16) future tense verbs.

5. The method of claim 5 wherein the psycho-linguistic cues are separated into categories, including increasing trend cues and decreasing trend cues and are analyzed according to presence in a category from within the categories.

6. A method for determining whether a text is deceptive, comprising:

establishing a set of cues;

determining if a cue in the set of cues is part of one of a sub-set of increasing trend cues and a sub-set of decreasing trend cues;

for increasing trend cues,

for each ith deception indicator cue value a_i, determining, via a computing device, whether a_i for a text is higher than an average value ā

_idc determined from evaluation of a deceptive text training data set;

when a_i>

ā

_idc labeling this deception indicator cue a_i as a match for the text and setting a deceptive coefficient for the ith cue, c_i, equal to 1;

determining, via a computing device, whether a_i is lower than the average value ā

_itr determined from evaluation of a truthful text training set;

when a_i<

ā

_itr labeling this deception indicator cue a_i as not a match for the text and setting the deception coefficient for the ith cue, c_i, equal to 0;

determining, via a computing device, whether a_i is between ā

_itr and ā

_idc;

computing, via a computing device, how close a_i is to ā

_idc and setting the deceptive coefficient c_i to a value between 0 to 1 accordingly;

for decreasing trend cues

for each ith deception indicator cue value, a_i determining, via a computing device, whether a_i for a text is lower than an average value ā

_idc;

labeling this deception indicator cue a_i as a match for the text and setting the deceptive coefficient for the ith cue, c_i, equal to 1;

determining, via a computing device, whether a_i is higher than the average value ā

_itr;

when a_i>

ā

_itr labeling this deception indicator cue a_i as not a match for the text and setting the deception coefficient for the ith cue, c_i, equal to 0;

determining, via a computing device, whether a_i for the text is between ā

_itr and ā

_idc,

computing, via a computing device, how close a_i is to ā

_itr and setting the deceptive coefficient c_i a value between 0 to 1 accordingly;

summing the deceptive coefficients c_i to obtain a deceptive value d; and

,

labeling the text as deceptive at least in part based upon the value d.

7. The method of claim 6 further comprising:

labeling the text as deceptive at least in part based on d being greater than a selected threshold value t.

8. The method of claim 6 wherein:

the sub-set of increasing trend cues includes cues where ā

_idc>

ā

_itr for the respective cue and the subset of decreasing trend cues included cues where ā

_idc<

ā

_itr for the respective cue.

9. The method of claim 6, further comprising:

weighting each c_i.

10. The method of claim 9 wherein the weighting is according to a simulated annealing algorithm.

11. A process for determining whether a text is deceptive, comprising:

establishing a set of cues;

evaluating, via a computing device, whether the text is deceptive based on a Markov chain evaluation of cue sequences in the text.

12. The process of claim 11 wherein the Markov chain evaluation comprises utilizing an evaluation of the probability of a transition from a first cue state to a second cue state for a plurality of cues defining a series of cues possibly occurring within a text containing m cues.

13. The method of claim 11 further comprising:

determining the number of words n in the text;

assigning a state to each word 1 . . . n in the text according to whether the word is within a particular cue category;

computing, via the computing device, the probabilities of n consecutive transitions, P_ndec and P_ntru, according to the equation;

14. The method of claim 13, further comprising:

labeling the text as deceptive where P_ndec>

P_ntru.

15. A method of determining whether a text is deceptive comprising:

selecting a desired detection rate α

;

selecting a false alarm probability β

;

setting a decision threshold A=(1−

β

)/α

;

setting a decision threshold B=β

/(1−

α

);

for a parameter being tested θ

, set a null hypothesis H₀;

θ

=θ

₀ and H₁;

θ

=θ

₁;

let a probability distribution of a random variable X=f(X,θ

₀) if H₀ is true and X=f(X,θ

₁) if H₁ is true;

determining, via a computing device, p_1n=f(X₁,θ

₁), f(X₂,θ

₁), . . . f(X_n, θ

₁);

determining, via a computing device, p_0n=f(X₁,θ

₀), f(X₂,θ

₀), . . . f(X_n,θ

₀);

designating H₁ as true if p_1n/p_0n≧

A; and

designating H₀ as true if p_1n/p_0n≦

B.

16. The method of claim 15 further comprising:

incrementing n to n+1 and repeating the steps of claim 15 if neither of p_1n/p_0n≧

A and p_1n/p_0n≦

B is true.