System and method for detecting fraudsters

1. A method for classifying a plurality of interactions, the method comprising, using a processor:

flagging one or more of the plurality of interactions based on one or more rules, wherein one or more of the rules describes an agent to be monitored;

obtaining a plurality of voiceprints of the plurality of flagged interactions including a set of benchmark voiceprints, wherein each voiceprint of the plurality of voiceprints represents a speaker participating in an interaction of the plurality of flagged interactions, and wherein the benchmark voiceprints are selected from one or more of the plurality of the voiceprints including a plurality of different speakers;

calculating, for each interaction of the plurality of flagged interactions, a plurality of scores, wherein each score of the plurality of scores is indicative of a similarity between the voiceprint of the interaction and one voiceprint of a set of benchmark voiceprints;

calculating, for each interaction of the plurality of flagged interactions, statistics of the scores;

finding candidate pairs of interactions based on similarity in the statistics of the scores of each interaction in the candidate pair;

for each candidate pair:

calculating a pair threshold, based on the statistics of the scores of the candidate pair;

calculating a pair score indicative of a similarity between the voiceprints associated with the interactions of the pair; and

determining that the pair pertains to a single cluster of interactions if the pair score is above the pair threshold; and

clustering one or more of the interactions of the plurality of flagged interactions into a single cluster based on one or more of the pairs that pertain to a single cluster.

2. The method of claim 1 , wherein the benchmark voiceprints are randomly selected from the plurality of voiceprints.

3. The method of claim 1 , wherein finding the candidate pairs of interactions comprises:

calculating similarity between the statistics of the scores; and

finding pairs for which the similarity is above a similarity threshold.

4. The method of claim 1 , wherein calculating statistics of the scores comprises calculating means and standard deviations, and wherein finding the candidate pairs of interactions comprises:

finding pairs of interactions for which the difference between the mean associated with each interaction is below a second threshold and the difference between the standard deviation associated with each interaction is below a third threshold.

5. The method of claim 1 , wherein calculating the pair threshold comprises:

finding, based on the statistics of the scores of the candidate pair, a value that is above a majority of the scores of the interactions of the pair.

6. The method of claim 1 , wherein calculating the pair threshold comprises:

finding the maximum between the mean plus Z times the standard deviation of the scores of each interaction of the pair, wherein Z is a number equal or larger than two.

7. The method of claim 1 , comprising:

clustering the interactions based on the pair that pertain to a single cluster by traversing the pairs that pertain to a single cluster; and for each pair:

if none of the interactions in the pair already pertain to a cluster, assigning the interactions in the pair to a new cluster;

if a first interaction in the pair pertains to a cluster and a second interaction in the pair does not pertain to a cluster, assigning the second interaction to the same cluster as the first interaction; and

if a first interaction in the pair pertains to a first cluster and a second interaction in the pair pertains to a second cluster, unifying the first cluster and the second cluster into a single cluster.

8. The method of claim 1 , wherein calculating statistics of the scores comprises calculating means and standard deviations.

9. The method of claim 1 , comprising filtering out one or more interactions from the detecting of pairs of similar speakers based on the statistics of the scores.

10. The method of claim 1 , comprising displaying information on the cluster, the information comprising a rank, the rank based on a likelihood that the cluster describes a single speaker.

11. A system for classifying a plurality of interactions, the system comprising:

a memory; and

a processor configured to:

flag one or more of the plurality of interactions based on one or more rules, wherein one or more of the rules describes an agent to be monitored;

obtain a plurality of voiceprints of the plurality of flagged interactions including a set of benchmark voiceprints, wherein each voiceprint of the plurality of voiceprints represents a speaker participating in an interaction of the plurality of flagged interactions, and wherein the benchmark voiceprints are selected from one or more of the plurality of the voiceprints including a plurality of different speakers;

calculate, for each interaction of the plurality of flagged interactions, a plurality of scores, wherein each score of the plurality of scores is indicative of a similarity between the voiceprint of the interaction and one voiceprint of a set of benchmark voiceprints;

calculate, for each interaction of the plurality of flagged interactions, statistics of the scores;

find candidate pairs of interactions based on similarity in the statistics of the scores of each interaction in the candidate pair;

for each candidate pair:

calculate a pair threshold, based on the statistics of the scores of the candidate pair;

calculate a pair score indicative of a similarity between the voiceprints associated with the interactions of the pair; and

determine that the pair pertains to a single cluster of interactions if the pair score is above the pair threshold; and

cluster one or more of the interactions of the plurality of flagged interactions into a single cluster based on one or more of the pairs that pertain to a single cluster.

12. The system of claim 11 , wherein the processor is configured to select the benchmark voiceprints randomly from the plurality of voiceprints.

13. The system of claim 11 , wherein the processor conlguired to find the candidate pairs of interactions by:

calculating similarity between the statistics of the scores; and

finding pairs for which the similarity is above a similarity threshold.

14. The system of claim 11 , wherein the processor configured to calculate statistics of the scores by calculating means and standard deviations, and to find the candidate pairs of interactions by:

finding pairs of interactions for which the difference between the mean associated with each interaction is below a second threshold and the difference between the standard deviation associated with each interaction is below a third threshold.

15. The system of claim 11 , wherein the processor configured to calculate the pair threshold by:

finding, based on the statistics of the scores of the candidate pair, a value that is above a majority of the scores of the interactions of the pair.

16. The system of claim 11 , wherein the processor configured to calculate the pair threshold by:

finding the maximum between the mean plus Z times the standard deviation of the scores of each interaction of the pair, wherein Z is a number equal or larger than two.

17. The system of claim 11 , the processor configured to:

cluster the interactions based on the pair that pertain to a single cluster by traversing the pairs that pertain to a single cluster; and for each pair:

if none of the interactions in the pair already pertain to a cluster, assign the interactions in the pair to a new cluster;

if a first interaction in the pair pertains to a cluster and a second interaction in the pair does not pertain to a cluster, assign the second interaction to the same cluster as the first interaction; and

if a first interaction in the pair pertains to a first cluster and a second interaction in the pair pertains to a second cluster, unify the first cluster and the second cluster into a single cluster.

18. The system of claim 11 , wherein the processor configured to calculate statistics of the scores by calculating means and standard deviations.