One or more non-transitory computer-readable media comprising computer executable instructions that, when executed, cause at least one processor to perform actions comprising:
- processing a group of customer support sessions for each time window of a sequence of time windows to determine if an event occurred in the group of customer support sessions, wherein the sequence of time windows comprises a first time window and a second time window;
- obtaining information about possible customer parameters, wherein each customer parameter relates to information about a customer;
- obtaining a first group of customer support sessions from the first time window;
- determining a first plurality of topics by processing text of the first group of customer support sessions with a neural network, wherein each customer support session of the first group is associated with a topic of the first plurality of topics;
- for each customer support session of the first group, selecting a subset of the possible customer parameters that correspond to the customer support session;
- computing first counts for the first group, wherein:
- each count of the first counts corresponds to (i) a topic of the first plurality of topics and (ii) a customer parameter of the possible customer parameters, and
- computing the first counts comprises processing a first customer support session of the first group by incrementing a subset of the first counts corresponding to (i) a topic of the first customer support session and (ii) a first subset of the possible customer parameters corresponding to the first customer support session;
- computing first baseline counts, wherein each baseline count of the first baseline counts corresponds to (i) a topic of the first plurality of topics and (ii) a customer parameter of the possible customer parameters;
- computing first event detection scores by processing the first counts and the first baseline counts, wherein each event detection score of the first event detection scores corresponds to (i) a topic of the first plurality of topics and (ii) a customer parameter of the possible customer parameters;
- determining that no events occurred by processing the first event detection scores;
- obtaining a second group of customer support sessions from the second time window;
- determining a second plurality of topics by processing text of the second group of customer support sessions with the neural network, wherein each customer support session of the second group is associated with a topic of the second plurality of topics;
- for each customer support session of the second group, selecting a subset of the possible customer parameters that correspond to the customer support session;
- computing second counts for the second group, wherein:
- each count of the second counts corresponds to (i) a topic of the second plurality of topics and (ii) a customer parameter of the possible customer parameters, and
- computing the second counts comprises processing a second customer support session of the second group by incrementing a subset of the second counts corresponding to (i) a topic of the second customer support session and (ii) a second subset of the possible customer parameters corresponding to the second customer support session;
- computing second baseline counts, wherein each baseline count of the second baseline counts corresponds to (i) a topic of the second plurality of topics and (ii) a customer parameter of the possible customer parameters;
- computing second event detection scores by processing the second counts and the second baseline counts, wherein each event detection score of the second event detection scores corresponds to (i) a topic of the second plurality of topics and (ii) a customer parameter of the possible customer parameters;
- and
- determining that a first event occurred by processing the second event detection scores.