Granted Patent B2

US 11,501,068 · App. 17/003,715 · Granted Nov 15, 2022

Inventors: Anish Parikh (Longmont, CO); Evan M. Caron (Houston, TX)

Assignee: FairWords, Inc.

G06F40/274G06F40/205G06F40/30H04L51/02H04L51/046H04L65/40H04L51/04H04L67/30

Abstract

Claims (82)

1. A method of preventing corporate- and legal compliance-violating messages by precluding keystrokes associated with non-compliant messages from reaching a target application via operations occurring between hardware and application layers of a computing device, the method comprising:

capturing a keystroke data signal from a text input hardware device of the computing device, the keystroke data signal representing at least a sentence or a phrase;

determining at an end of the sentence or the phrase, via a processor, if the sentence or phrase is non-compliant by:

checking the sentence or phrase for any typos;

generating a first list of suggested typo fixes;

generating a second list of suggested sentences or phrases each incorporating one of the suggested typo fixes from the first list of suggested typo fixes;

determining risk scores for suggested sentences or phrases in the second list of suggested sentences or phrases;

aggregating the risk scores for the suggested sentences to form an aggregated risk score for the sentence or phrase;

comparing the aggregated risk score to a compliance violation threshold; and

if the aggregated risk score exceeds the compliance violation threshold, then deeming the sentence or phrase non-compliant and, at a layer between a hardware and an application layer, precluding at least a portion of the keystroke data signal representing at least the sentence or the phrase from reaching the target application in the application layer.

2. The method of claim 1 , wherein the keystroke data signal represents at least a first word and the method further comprising determining if the first word is non-compliant, and if so, then precluding at least a portion of the keystroke data signal representing at least the first word from reaching the target application.

3. The method of claim 2 , wherein the determining if the first word is non-compliant comprises comparing the first word to a list of non-compliant words.

4. The method of claim 2 , wherein the determining if the first word is non-compliant comprises:

creating postfixes;

comparing the postfixes to a spellchecker dictionary, and if one of the postfixes matches one or more words in the spellchecker dictionary, then lemmatizing the one of the postfixes and adding the one of the postfixes to a first list of lemmas, and if the one of the postfixes does not match the one or more words in the spellchecker dictionary, then generating the first list of suggested typo fixes, lemmatizing each suggested typo fix in the first list of suggested typo fixes to form lemma typo fix suggestions, and adding the lemma typo fix suggestions to the first list of lemmas; and

calculating a Damerau-Levenshtein edit (DL) distance for each lemma in the first list of lemmas and assigning non-compliant tags if lemmas associated with the DL distance are below a threshold.

5. The method of claim 4 , wherein the generating the second list comprises:

generating a list of modified sentences or phrases based on the sentence or phrase as modified by suggested typo fixes from the first list of suggested typo fixes;

accessing one or more phrase templates;

creating a second list of lemmas from the one or more phrase templates;

ordering the list of modified sentences or phrases based on an occurrence of words within each suggested sentence or phrase having a match in the second list of lemmas to form an ordered list of suggested sentences or phrases; and

selecting a top N of the ordered list of suggested sentences or phrases as the second list of suggested sentences or phrases.

6. The method of claim 5 , wherein the determining the risk score comprises:

for each of the suggested sentences or phrases in the second list of suggested sentences or phrases,

tokenizing the suggested sentence or phrase into tokens;

assigning each of the tokens a part of speech to form parts-of-speech-tagged tokens;

comparing the parts-of-speech-tagged tokens to the one or more phrase templates;

accumulating a score for the suggested sentence or phrase based on matches between parts-of-speech-tagged tokens and parts of the one or more phrase templates; and

returning the score as the risk score for the suggested sentence or phrase.

7. The method of claim 6 , further comprising:

determining a semantic graph for the suggested sentence or phrase;

comparing the semantic graph to the one or more phrase templates; and

adding a higher score to the risk score if there is a match to the semantic graph and adding a lower score to the risk score if there is not a match to the semantic graph.

8. The method of claim 1 , further comprising, displaying a pop-up window on a graphical interface when a non-compliant sentence or phrase is identified.

9. The method of claim 1 , further comprising, removing non-compliant parts of the sentence or the phrase from a graphical interface of the computing device and from a computer memory of the computing device.

10. A method of preventing corporate- and legal compliance-violating messages by precluding keystrokes associated with non-compliant messages from reaching a target application via operations occurring between hardware and application layers of a computing device, the method comprising:

capturing a keystroke data signal from a text input hardware device of the computing device, the keystroke data signal representing at least a sentence or phrase;

determining at an end of the sentence or the phrase, via a processor, if the sentence or phrase is non-compliant by:

determining an aggregated risk score for each sentence or phrase by identifying possible typo-based variations on the sentence or phrase, scoring the possible typo-based variations, and aggregating scores for the possible typo-based variations;

comparing the aggregated risk score to a compliance violation threshold; and

if the aggregated risk score exceeds the compliance violation threshold, then deeming the sentence or phrase non-compliant and, at a layer between a hardware and an application layer, precluding at least a portion of the keystroke data signal representing at least the sentence or the phrase from reaching the target application in the application layer.

11. The method of claim 10 , wherein the keystroke data signal represents at least a first word and the method further comprising determining if the first word is non-compliant, and if so, then precluding at least a portion of the keystroke data signal representing at least the first word from reaching the target application.

12. The method of claim 11 , wherein the determining if the first word is non-compliant comprises comparing the first word to a list of non-compliant words.

13. The method of claim 11 , wherein the determining if the first word is non-compliant comprises:

creating postfixes;

comparing the postfixes to a spellchecker dictionary, and if one of the postfixes matches one or more words in the spellchecker dictionary, then lemmatizing the one of the postfixes and adding the one of the postfixes to a first list of lemmas, and if the one of the postfixes does not match the one or more words in the spellchecker dictionary, then generating the first list of suggested typo fixes, lemmatizing each suggested typo fix in the first list of suggested typo fixes to form lemma typo fix suggestions, and adding the lemma typo fix suggestions to the first list of lemmas; and

calculating a Damerau-Levenshtein edit (DL) distance for each lemma in the first list of lemmas and assigning non-compliant tags if lemmas associated with the DL distance are below a threshold.

14. The method of claim 13 , wherein the determining the aggregated risk score comprises:

checking the sentence or phrase for any typos;

generating a first list of suggested typo fixes; and

generating a second list of suggested sentences or phrases each incorporating one of the suggested typo fixes from the first list of suggested typo fixes by:

generating a list of modified sentences or phrases based on the sentence or phrase as modified by suggested typo fixes from the first list of suggested typo fixes;

accessing one or more phrase templates;

creating a second list of lemmas from the one or more phrase templates;

ordering the list of modified sentences or phrases based on an occurrence of words within each suggested sentence or phrase having a match in the second list of lemmas to form an ordered list of suggested sentences or phrases; and

selecting a top N of the ordered list of suggested sentences or phrases as the second list of suggested sentences or phrases.

15. The method of claim 14 , wherein the determining the aggregated risk score further comprises:

for each of the suggested sentences or phrases in the second list of suggested sentences or phrases,

tokenizing the suggested sentence or phrase into tokens;

assigning each of the tokens a part of speech to form parts-of-speech-tagged tokens;

comparing the parts-of-speech-tagged tokens to the one or more phrase templates;

accumulating a score for the suggested sentence or phrase based on matches between parts-of-speech-tagged tokens and parts of the one or more phrase templates; and

returning the score as the risk score for the suggested sentence or phrase.

16. The method of claim 15 , further comprising:

determining a semantic graph for the suggested sentence or phrase;

comparing the semantic graph to the one or more phrase templates; and

adding a higher score to the risk score if there is a match to the semantic graph and adding a lower score to the risk score if there is not a match to the semantic graph.

17. The method of claim 10 , further comprising, displaying a pop-up window on a user display when a non-compliant sentence or phrase is identified.

18. The method of claim 10 , further comprising, removing non-compliant parts of the sentence or the phrase from a graphical interface of the computing device and from a temporary memory of the computing device.

19. A non-transitory, tangible computer readable storage medium, encoded with processor readable instructions to perform a method of preventing corporate- and legal compliance-violating messages by precluding keystrokes associated with non-compliant messages from reaching a target application via operations occurring between hardware and application layers of a computing device, the method comprising:

capturing a keystroke data signal from a text input hardware device of the computing device, the keystroke data signal representing at least a sentence or phrase; and

determining at an end of the sentence or the phrase, via a processor, if the sentence or phrase is non-compliant by:

determining an aggregated risk score for each sentence or phrase by identifying possible typo-based variations on the sentence or phrase, scoring the possible typo-based variations, and aggregating scores for the possible typo-based variations;

comparing the aggregated risk score to a compliance violation threshold; and

if the aggregated risk score exceeds the compliance violation threshold, then deeming the sentence or phrase non-compliant and, at a layer between a hardware and an application layer, precluding at least a portion of the keystroke data signal representing at least the sentence or the phrase from reaching the target application in the application layer.

20. The non-transitory, tangible computer readable storage medium of claim 19 , wherein the determining the risk score comprises:

for each of the suggested sentences or phrases in the second list of suggested sentences or phrases,

tokenizing the suggested sentence or phrase into tokens;

assigning each of the tokens a part of speech to form parts-of-speech-tagged tokens;

comparing the parts-of-speech-tagged tokens to one or more phrase templates;

accumulating a score for the suggested sentence or phrase based on matches between parts-of-speech-tagged tokens and parts of the one or more phrase templates; and

returning the score as the risk score for the suggested sentence or phrase.

Assignments (3)

Continuity (10)

Cited By (1)

US 12,282,739