Method and system for fraud clustering by content and biometrics analysis
A computer-implemented method for proactive fraudster exposure in a customer service center according to content analysis and voice biometrics analysis, is provided herein. The computer-implemented method includes: (i) collecting call interaction; (ii) storing the collected call interactions; (iii) performing a first type analysis to cluster the call interactions into ranked clusters and storing the ranked clusters in a clusters database; (iv) performing a second type analysis on a predefined amount of the highest ranked clusters, into ranked clusters and storing the ranked clusters; the first type analysis is a content analysis and the second type analysis is a voice biometrics analysis, or vice versa, (v) enabling a user to repeat steps (iii) and (iv); (vi) retrieving from the ranked clusters, a list of fraudsters; and transmitting the list of potential fraudsters to an application to display to a user said list of potential fraudsters via a display unit.
1. A computer-implemented method for proactive fraudster exposure in a customer service center according to content analysis and voice biometrics analysis of a call interaction, the computer-implemented method comprising:
(a) collecting call interactions from a database of recorded calls in a customer service channel, wherein the collecting is based on predefined rules by a calls collection engine,
(b) storing the collected call interactions in an interactions database;
(c) performing a first type analysis to cluster the call interactions in the interactions database into first type ranked clusters and storing the first type ranked clusters in a clusters database;
(d) performing a second type analysis on a predefined amount of highest ranked clusters of the first type ranked clusters, into second y ranked clusters and storing the second type ranked clusters in the clusters database;
wherein the first type analysis is a content analysis of a call interaction and the second type analysis is a voice biometrics analysis of the call interaction, or vice versa,
enabling a user to repeat steps (c) and (d);
retrieving from the second type ranked clusters in the clusters database, a list of potential fraudsters;
transmitting the list of potential fraudsters to an application to display to a user said list of potential fraudsters via a display unit, thus enabling said user to review said list of potential fraudsters and to add fraudsters from the list to a watchlist database.
2. The computer-implemented method of claim 1 , wherein the content analysis of the call interaction comprising:
performing Automatic Speech Recognition (ASR) for each call interaction to yield a textual transcript thereof;
operating a clustering algorithm over each textual transcript to cluster phrases in the textual transcript into clusters according to topics; and
identifying clusters with topics having predefined suspicious phrases and ranking the identified clusters.
3. The computer-implemented method of claim 2 , wherein the identifying is performed according to at least one of the following algorithms: graph theory, machine learning, statistics, linguistic, deep learning, hybrid, Term Frequency-Inverse Document Frequency (TF-IDF) and parallel word importance estimator.
4. The computer-implemented method of claim 2 , wherein, the content analysis of the call interaction further comprising:
representing a predefined amount of highest ranked phrases as nodes in a connected graph;
calculating phrase statistics to suggest to a user a connection between the phrases according to the graph; and
enabling a user to select a specific phrase in the graph for biometrics analysis of related call interactions.
5. The computer-implemented method of claim 1 , wherein the voice biometrics analysis of the call interaction is performed by a Proactive Fraud Exposure (PFE) engine, wherein the PFE is performing clustering.
6. The computer-implemented method of claim 5 , wherein the PFE engine is performing the clustering according to at least one of: K-means, K-centroids, Clustering Using REpresentatives (CURE), and Agglomerative Hierarchal Clustering (AHC) algorithms.
7. The computer-implemented method of claim 2 , wherein the content analysis of the call interaction further includes stemming to transform phrases into a predefined form.
8. A computerized-system for proactive fraudster exposure in a customer service center according to content analysis and voice biometrics analysis of a call interaction, the computerized-system comprising:
a database of recorded calls;
an interactions database;
a clusters database,
a watchlist database;
a memory to store the database of recorded calls, the interactions database, the clusters database and the watchlist database;
a display unit;
and
a processor, said processor is configured to:
(a) collect call interactions from the database of recorded calls in a customer service channel, wherein the collecting is based on predefined rules by a calls collection engine,
(b) store the collected call interactions in the interactions database;
(c) perform a first type analysis to cluster the call interactions in the interactions database into first type ranked clusters and to store the first type ranked clusters in the clusters database;
(d) perform a second type analysis on a predefined amount of highest ranked clusters, into second type ranked clusters and to store the second type ranked clusters in the clusters database;
wherein the first type analysis is a content analysis of a call interaction and the second type analysis is a voice biometrics analysis of the call interaction, or vice versa, enable a user to repeat steps (c) and (d);
retrieve from the second type ranked clusters in the clusters database, a list of potential fraudsters;
transmit the list of potential fraudsters to an application to display to a user said list of potential fraudsters via a display unit, thus, to enable said user to review said list of potential fraudsters and to add fraudsters from the list to the watchlist database.
9. The computerized-system of claim 8 , wherein the content analysis of the call interaction comprising:
performing Automatic Speech Recognition (ASR) for each call interaction to yield a textual transcript thereof;
operating a clustering algorithm over each textual transcript to cluster phrases in the textual transcript into clusters according to topics; and
identifying clusters with topics having predefined suspicious phrases and ranking the identified clusters.
10. The computerized-system of claim 9 , wherein the identifying is performed according to at least one of the following algorithms: graph theory, machine learning, statistics, linguistic, deep learning, hybrid, Term Frequency-Inverse Document Frequency (TF-IDF) and parallel word importance estimator.
11. The computerized-system of claim 9 , wherein, the content analysis of the call interaction further comprising:
representing a predefined amount of highest ranked phrases as nodes in a connected graph;
calculating phrase statistics to suggest to a user a connection between the phrases according to the graph; and
enabling a user to select a specific phrase in the graph for biometrics analysis of related call interactions.
12. The computerized-system of claim 8 , wherein the voice biometrics analysis of the call interaction is performed by a Proactive Fraud Exposure (PFE) engine, wherein the PFE is performing clustering.
13. The computerized-system of claim 12 , wherein the PFE engine is performing the clustering according to at least one of: K-means, K-centroids, Clustering Using REpresentatives (CURE), and Agglomerative Hierarchal Clustering (AHC) algorithms.
14. The computerized-system of claim 9 , wherein the content analysis of the call interaction further includes stemming to transform phrases into a predefined form.