| CPC G06Q 40/12 (2013.12) [G06N 5/04 (2013.01); G06N 20/00 (2019.01)] | 10 Claims |
|
1. A method comprising:
extracting, by a transaction machine learning model at a first stage comprising a number of pre-trained encoders, a plurality of sparse raw features from a transaction record of a transaction, wherein the transaction machine learning model is a general model trained on transaction data from a plurality of entities to convert sparse features to dense features, wherein the transaction record is for a particular entity;
encoding, by the number of pre-trained encoders in the first stage, the plurality of sparse raw features into a transaction vector comprising a plurality of dense features;
classifying, by an adapter model at a second stage that is a machine learning model trained, for the particular entity, on historical transactions in a customized chart of accounts of the particular entity, the transaction vector into the customized chart of accounts using the plurality of dense features to generate adapter model output;
processing the adapter model output to select an account identifier that corresponds to the transaction record and to an account of the customized chart of accounts; and
presenting the account identifier for the transaction record.
|