| CPC G06Q 40/12 (2013.12) [G06N 3/048 (2023.01); G06N 3/10 (2013.01)] | 19 Claims |
|
1. A computer-implemented method comprising:
extracting for each of a plurality of submitted expenses an expense type and corresponding comment text entered by a respective user of a plurality of users to generate training data comprising the extracted expense type and corresponding comment text for each of the plurality of submitted expenses;
generating a customized machine learning model by:
adding, to a pretrained machine learning model, a first customized layer to prevent overfitting and a second customized layer comprising a classifier to generate output; and
training the pretrained machine learning model with the added first customized layer and second customized layer using the training data comprising the extracted expense type and corresponding comment text entered by a respective user of a plurality of users for each of the plurality of submitted expenses to generate a trained machine learning model configured to predict, in real time, an expense type for a given expense entry by a respective user;
detecting user entry in a user interface on a respective computing device of comment text associated with an expense for each of a plurality of entered expenses from a plurality of computing devices; and
automatically updating, in real time from detecting user entry of comment text associated with an expense for each of the plurality of entered expenses from the plurality of computing devices, incorrect expense types by performing operations comprising:
generating, in real time, a predicted expense type for each entered expense and a confidence value indicating a likelihood that the predicted expense type is correct, by analyzing, by the trained machine learning model, the comment text of each user entry of the plurality of entered expenses from the plurality of computing devices;
comparing the predicted expense type to a corresponding expense type entered for the expense for each of the plurality of entered expenses to determine whether the predicted expense type matches the expense type entered for the expense;
determining that a subset of the plurality of entered expenses have expense types entered that do not match a corresponding predicted expense type and have a predicted expense type with a confidence value over a threshold confidence value;
causing each expense type of the subset of the plurality of entered expenses to be automatically updated with the predicted expense type; and
causing display, on each user interface on the respective computing device associated with each of the plurality of entered expenses, of a notification indicating that the expense type entered via user entry was incorrect.
|