Protecting sensitive data using conversational history
Methods and systems for protecting sensitive data using conversational history are described herein. An enterprise data validation server may receive conversation snippets and create a topic model. The enterprise data validation server may detect a message is being sent from a first user to a second user, determine a topic distribution between the first user and the second user, and a topic distribution of the message. The enterprise data validation server may determine a bias value associated with the message by comparing the topic distribution of the message and the topic distribution between the first user and the second user. Accordingly, based on a determination that the bias value exceeds a threshold, the enterprise data validation server send an alert containing a warning message.
1. A method comprising:
at an enterprise data validation server comprising at least one processor, memory, and a communication interface:
receiving, via the communication interface, conversation snippets associated with a plurality of applications in a virtual computing environment;
creating, based on the conversation snippets, a topic model using machine learning modeling;
detecting, via the communication interface, a message is being sent from a first user to a second user;
determining, based on the topic model, a topic distribution between the first user and the second user;
determining a topic distribution of the message;
determining a bias value associated with the message by comparing the topic distribution of the message and the topic distribution between the first user and the second user;
adjusting the bias value based on sensitive information being in the message; and
based on a determination that the bias value exceeds a threshold, sending an alert to the first user to prevent the message being sent to the second user.
2. The method of claim 1 , wherein the machine learning modeling comprises latent Dirichlet allocation (LDA) modeling.
3. The method of claim 1 , further comprising:
prior to receiving the conversation snippets, capturing a HTTP message associated with the conversation snippets from a browser of an application of the plurality of the applications.
4. The method of claim 1 , further comprising:
prior to receiving the conversation snippets, capturing the conversation snippets from native codes of an application of the plurality of the applications.
5. The method of claim 1 , further comprising:
after receiving the conversation snippets, storing the conversation snippets in a corpus database.
6. The method of claim 1 , further comprising:
after creating the topic model, generating the topic distribution associated with conversations associated with a plurality of users; and
storing the topic distribution in a topic distribution database.
7. The method of claim 1 , further comprising:
using the topic distribution and a topic number as inputs for the machine learning modeling.
8. The method of claim 1 , further comprising:
tuning the topic model to identify an appropriate topic number.
9. The method of claim 1 , further comprising:
applying the topic model to a production environment; and
tuning the topic model based on user feedback.
10. The method of claim 1 , further comprising:
sending the alert indicating that the sensitive information in the message is to be disclosed to the second user; and
blocking the message from being sent to the second user based on a negative feedback from the first user.
11. The method of claim 1 , further comprising:
sending the alert indicating that the sensitive information in the message is to be disclosed to the second user; and
permitting the message being sent to the second user based on a positive feedback from the first user.
12. An enterprise data validation server comprising:
at least one processor;
a communication interface;
memory storing instructions that, when executed by the at least one processor, cause the enterprise data validation server to:
receive, via the communication interface, conversation snippets associated with a plurality of applications in a virtual computing environment;
create, based on the conversation snippets, a topic model using machine learning modeling;
detect, via the communication interface, a message is being sent from a first user to a second user;
determine, based on the topic model, a topic distribution between the first user and the second user;
determine a topic distribution of the message;
determine a bias value associated with the message by comparing the topic distribution of the message and the topic distribution between the first user and the second user;
adjust the bias value based on sensitive information being in the message; and
based on a determination that the bias value exceeds a threshold, send an alert to the first user to prevent the message from being sent to the second user.
13. The enterprise data validation server of claim 12 , wherein the memory stores additional instructions that, when executed by the at least one processor, cause the enterprise data validation server to:
create the topic model using latent Dirichlet allocation (LDA) modeling.
14. The enterprise data validation server of claim 12 , wherein the memory stores additional instructions that, when executed by the at least one processor, cause the enterprise data validation server to:
prior to receiving the conversation snippets, capture HTTP message associated with the conversation snippets from a browser of an application of the plurality of the applications.
15. The enterprise data validation server of claim 12 , wherein the memory stores additional instructions that, when executed by the at least one processor, cause the enterprise data validation server to:
prior to receiving the conversation snippets, capture the conversation snippets from native codes of an application of the plurality of the applications.
16. The enterprise data validation server of claim 12 , wherein the memory stores additional instructions that, when executed by the at least one processor, cause the enterprise data validation server to:
after creating the topic model, generate the topic distribution associated with conversations associated with a plurality of users; and
storing the topic distribution in a topic distribution database.
17. The enterprise data validation server of claim 12 , wherein the memory stores additional instructions that, when executed by the at least one processor, cause the enterprise data validation server to:
using the topic distribution and a topic number as inputs for the machine learning modeling.
18. The enterprise data validation server of claim 12 , wherein the memory stores additional instructions that, when executed by the at least one processor, cause the enterprise data validation server to:
tune the topic model to identify an appropriate topic number.
19. One or more non-transitory computer-readable media storing instructions that, when executed by a computing platform comprising at least one processor, memory, and a communication interface, cause the computing platform to:
receive, via the communication interface, conversation snippets associated with a plurality of applications in a virtual computing environment;
create, based on the conversation snippets, a topic model using machine learning modeling;
detect, via the communication interface, a message is being sent from a first user to a second user;
determine, based on the topic model, a topic distribution between the first user and the second user;
determine a topic distribution of the message;
determine a bias value associated with the message by comparing the topic distribution of the message and the topic distribution between the first user and the second user;
adjust the bias value based on sensitive information being in the message; and
based on a determination that the bias value exceeds a threshold, send an alert to the first user to prevent the message being sent to the second user.