US 11,870,804 B2
Automated learning and detection of web bot transactions using deep learning
Sreenath Kurupati, Santa Clara, CA (US)
Assigned to Akamai Technologies, Inc., Cambridge, MA (US)
Filed by Akamai Technologies Inc., Cambridge, MA (US)
Filed on Aug. 1, 2019, as Appl. No. 16/529,408.
Prior Publication US 2021/0037048 A1, Feb. 4, 2021
Int. Cl. H04L 9/40 (2022.01); G06N 3/08 (2023.01)
CPC H04L 63/1458 (2013.01) [G06N 3/08 (2013.01); H04L 63/1466 (2013.01)] 12 Claims
OG exemplary drawing
 
1. A method to detect and distinguish traffic in a network operating environment, comprising:
generating a neural network model by:
receiving a first set of data representing current transactions;
receiving a second set of data representing transactions previously verified as being associated with human traffic;
artificially labeling each transaction in the first set of data with a first score indicative of a bot whether the transaction is a bot or a human;
labeling each transaction in the second set of data with a second score indicative of a human;
training a neural network using the first and second sets of data and the first and second scores as follows:
for each of a set of training iterations beginning with a first iteration:
determining whether the neural network converges;
responsive to a determination that the neural network does not converge, pruning one or more transactions from the first set of data that, based on a given threshold, cannot be discriminated from transactions in the second set of data;
determining whether the first set of data has sufficient samples;
responsive to a determination that the first set of data has sufficient samples, initiating a next iteration;
upon a determination that the neural network converges, outputting the neural network model; and
using the neural network model to discriminate traffic in the network operating environment as being either bot or human.