Machine learning powered authentication challenges
Systems and methods are disclosed for automatically predicting a risk score of a user login attempt by receiving a user login attempt and generating a login feature vector associated with the user login attempt. The systems and methods further train a machine learning technique to establish a relationship between the login feature vector and the risk score. The trained machine learning technique is applied to new user login attempts to predict a risk score associated with the login attempt and issue an authentication challenge to the user if the risk score exceeds a predetermined threshold value.
1. A computer-implemented method comprising:
receiving, by one or more processors, a plurality of user login attempts, each user login attempt associated with a user of a client device and a login source;
randomly sampling a portion of the plurality of user login attempts for each of the randomly sampled user login attempts:
generating a login feature vector associated with the user login attempt, the login feature vector comprising a plurality of user-context features that represent a probability of malicious software attacks;
associating the plurality of user-context features with a plurality of respective risk values;
determining a risk score associated with the user login attempt based on the plurality of respective risk values using a trained machine learning model;
determining that the risk score exceeds a predetermined threshold value;
in response to the determination that the risk score exceeds the predetermined threshold value, issuing an authentication challenge to the user; and
associating a challenge response label with the login feature vector based on a user response to the authentication challenge.
2. The computer-implemented method of claim 1 , further comprising:
determining, by the one or more processors, that the user satisfied the authentication challenge; and
in response to the determining that the user satisfied the authentication challenge, granting the user access to an online social platform.
3. The computer-implemented method of claim 1 , wherein the user-context features comprise a probability that the user can successfully login from a given Internet Protocol (IP) address.
4. The computer-implemented method of claim 1 , wherein the login feature vector further comprises one or both of login-context features that represent a probability of malicious software attacks associated with the login source.
5. The computer-implemented method of claim 1 , further comprising training a machine learning model, wherein training the machine learning model comprises:
receiving a plurality of user login attempts;
generating a login feature vector for each user login attempt of the plurality of user login attempts;
issuing a test authentication challenge to each of the users associated with the plurality of user login attempts;
determining a test authentication challenge result for each test authentication challenge; and
storing each test authentication challenge result and the respective login feature vector as a data pair in a data store.
6. The computer-implemented method of claim 5 , wherein a first portion of data pairs represent high-risk user login attempts and a second portion of data pairs represent low-risk user login attempts.
7. The computer-implemented method of claim 1 , wherein the predetermined threshold value represents a probability that the user login attempt is a malicious login attempt.
8. The computer-implemented method of claim 1 , further comprising:
causing the authentication challenge to be displayed on the user device.
9. A system comprising:
a memory: and
at least one hardware processor coupled to the memory and comprising instructions that cause the system to perform operations comprising:
receiving, by one or more processors, a plurality of user login attempts, each user login attempt associated with a user of a client device and a login source;
randomly sampling a portion of the plurality of user login attempts
for each of the randomly sampled user login attempts:
generating a login feature vector associated with the user login attempt, the login feature vector comprising a plurality of user-context features that represent a probability of malicious software attacks;
associating the plurality of user-context features with a plurality of respective risk values;
determining a risk score associated with the user login attempt based on the plurality of respective risk values using a trained machine learning model;
determining that the risk score exceeds a predetermined threshold value;
in response to the determination that the risk score exceeds the predetermined threshold value, issuing an authentication challenge to the user; and
associating a challenge response label with the login feature vector based on a user response to the authentication challenge.
10. The system of claim 9 , further comprising:
determining, by the one or more processors, that the user satisfied the authentication challenge; and
in response to the determining that the user satisfied the authentication challenge, granting the user access to an online social platform.
11. The system of claim 9 , wherein the user-context features comprise a probability that the user can successfully login from a given Internet Protocol (IP) address.
12. The system of claim 9 , wherein the login feature vector further comprises one or both of login-context features that represent a probability of malicious software attacks associated with the login source.
13. The system of claim 9 , further comprising training a machine learning model, wherein training the machine learning model comprises:
receiving a plurality of user login attempts;
generating a login feature vector for each user login attempt of the plurality of user login attempts;
issuing a test authentication challenge to each of the users associated with the plurality of user login attempts;
determining a test authentication challenge result for each test authentication challenge; and
storing each test authentication challenge result and the respective login feature vector as a data pair in a data store.
14. The system of claim 13 , wherein a first portion of data pairs represent high-risk user login attempts and a second portion of data pairs represent low-risk user login attempts.
15. The system of claim 9 , wherein the predetermined threshold value represents a probability that the user login attempt is a malicious login attempt.
16. The system of claim 9 , further comprising updating the determined risk score based on the login feature vector and the challenge response label associated with the login feature vector and the associated.
17. The system of claim 9 , wherein associating the plurality of user-context features with the plurality of respective risk values comprises associating the plurality of user-context features with a plurality of respective predefined risk values based on a predefined set of rules.
18. A non-transitory machine-readable storage medium comprising instructions that, when executed by one or more processors of a machine cause the machine to perform operations comprising:
receiving, by one or more processors, a plurality of user login attempts, each user login attempt associated with a user of a client device and a login source;
randomly sampling a portion of the plurality of user login attempts
for each of the randomly sampled user login attempts:
generating a login feature vector associated with the user login attempt, the login feature vector comprising a plurality of user-context features that represent a probability of malicious software attacks;
associating the plurality of user-context features with a plurality of respective risk values;
determining a risk score associated with the user login attempt based on the plurality of respective risk values using a trained machine learning model;
determining that the risk score exceeds a predetermined threshold value;
in response to the determination that the risk score exceeds the predetermined threshold value, issuing an authentication challenge to the user; and
associating a challenge response label with the login feature vector based on a user response to the authentication challenge.
19. The non-transitory machine-readable storage medium of claim 18 , further comprising:
determining, by the one or more processors, that the user satisfied the authentication challenge; and
in response to the determining that the user satisfied the authentication challenge, granting the user access to an online social platform.
20. The non-transitory machine-readable storage medium of claim 18 , wherein the user-context features comprise a probability that the user can successfully login from a given Internet Protocol (IP) address.