Bot-based honeypot poison resilient data collection
One embodiment provides a method comprising answering one or more incoming phone calls received at one or more pre-specified phone numbers utilizing a bot. The bot is configured to engage in a conversation with a caller initiating an incoming phone call utilizing a voice recording that impersonates a human being. The method further comprises recording each conversation the bot engages in, and classifying each recorded conversation as one of poison data or truthful training data based on content of the recorded conversation and one or more learned detection models for detecting poisoned data.
1. A method comprising:
answering one or more incoming phone calls received at one or more pre-specified phone numbers utilizing a bot, wherein the bot is configured to engage in a conversation with a caller who initiates an incoming phone call by utilizing a voice recording that impersonates an individual belonging to a type of demographic targeted by the caller and dynamically conversing with the caller based on a conversation template including one or more learned conversation patterns corresponding to the individual belonging to the type of demographic targeted by the caller;
recording each conversation the bot engages in; and
collecting training data for training a system in detecting unwanted phone solicitations, wherein the collecting comprises classifying each recorded conversation as one of poison data or truthful training data based on content of the recorded conversation and one or more learned detection models for detecting poisoned data, and the training data each comprises each recorded conversation classified as truthful training data.
2. The method of claim 1 , wherein each recorded conversation classified as truthful training data represents poison resilient training data available for use as the training data in training the system for filtering unwanted phone solicitations.
3. The method of claim 1 , wherein the bot is configured to select the voice recording and the conversation template from different voice recordings and different conversation templates corresponding to different individuals belonging to different types of demographic.
4. The method of claim 1 , further comprising:
for each recorded conversation, applying a speech-to-text process to the recorded conversation to obtain a transcript of the recorded conversation.
5. The method of claim 1 , wherein the one or more pre-specified phone numbers comprise one or more unused phone numbers.
6. The method of claim 1 , wherein use of the voice recording and the conversation template increases a likelihood the caller perceives the bot as a human being and makes a pitch to the bot.
7. The method of claim 1 , further comprising:
determining an identity of a caller initiating an incoming phone call by comparing voice patterns of the caller against one or more voice samples of one or more known scammers or telemarketers.
8. The method of claim 1 , wherein the one or more learned detection models comprise at least one of the following: data point inspection model, normality model, a set of rotatable classifiers, or a cross-source consistency checking model.
9. A system comprising:
at least one processor; and
a non-transitory processor-readable memory device storing instructions that when executed by the at least one processor causes the at least one processor to perform operations including:
answering one or more incoming phone calls received at one or more pre-specified phone numbers utilizing a bot, wherein the bot is configured to engage in a conversation with a caller who initiates an incoming phone call by utilizing a voice recording that impersonates an individual belonging to a type of demographic targeted by the caller and dynamically conversing with the caller based on a conversation template including one or more learned conversation patterns corresponding to the individual belonging to the type of demographic targeted by the caller;
recording each conversation the bot engages in; and
collecting training data for training a system in detecting unwanted phone solicitations, wherein the collecting comprises classifying each recorded conversation as one of poison data or truthful training data based on content of the recorded conversation and one or more learned detection models for detecting poisoned data, and the training data each comprises each recorded conversation classified as truthful training data.
10. The system of claim 9 , wherein each recorded conversation classified as truthful training data represents poison resilient training data available for use as the training data in training the system for filtering unwanted phone solicitations.
11. The system of claim 9 , wherein the bot is configured to select the voice recording and the conversation template from different voice recordings and different conversation templates corresponding to different individuals belonging to different types of demographic.
12. The system of claim 9 , the operations further comprising:
for each recorded conversation, applying a speech-to-text process to the recorded conversation to obtain a transcript of the recorded conversation.
13. The system of claim 9 , wherein the one or more pre-specified phone numbers comprise one or more unused phone numbers.
14. The system of claim 9 , wherein use of the voice recording and the conversation template increases a likelihood the caller perceives the bot as a human being and makes a pitch to the bot.
15. The system of claim 9 , the operations further comprising:
determining an identity of a caller initiating an incoming phone call by comparing voice patterns of the caller against one or more voice samples of one or more known scammers or telemarketers.
16. The system of claim 9 , wherein the one or more learned detection models comprise at least one of the following: data point inspection model, normality model, a set of rotatable classifiers, or a cross-source consistency checking model.
17. A computer program product comprising a computer-readable hardware storage medium having program code embodied therewith, the program code being executable by a computer to implement a method comprising:
answering one or more incoming phone calls received at one or more pre-specified phone numbers utilizing a bot, wherein the bot is configured to engage in a conversation with a caller who initiates an incoming phone call by utilizing a voice recording that impersonates an individual belonging to a type of demographic targeted by the caller and dynamically conversing with the caller based on a conversation template including one or more learned conversation patterns corresponding to the individual belonging to the type of demographic targeted by the caller;
recording each conversation the bot engages in; and
collecting training data for training a system in detecting unwanted phone solicitations, wherein the collecting comprises classifying each recorded conversation as one of poison data or truthful training data based on content of the recorded conversation and one or more learned detection models for detecting poisoned data, and the training data each comprises each recorded conversation classified as truthful training data.
18. The computer program product of claim 17 , wherein each recorded conversation classified as truthful training data represents poison resilient training data available for use as the training data in training the system for filtering unwanted phone solicitations.
19. The computer program product of claim 17 , wherein the bot is configured to select the voice recording and the conversation template from different voice recordings and different conversation templates corresponding to different individuals belonging to different types of demographic.
20. The computer program product of claim 17 , the method further comprising:
for each recorded conversation, applying a speech-to-text process to the recorded conversation to obtain a transcript of the recorded conversation.