CPC G06Q 40/12 (2013.12) [G06N 3/04 (2013.01); G06N 3/08 (2013.01); G06Q 30/018 (2013.01)] | 20 Claims |
1. A method comprising: training a comparison model without classifying inputs to the comparison model, wherein the comparison model comprises one or more neural network layers, wherein the training uses a processor that compares transaction record pairs to generate comparison scores, wherein a comparison score measures similarity of a transaction record pairs of the transaction record pair;
executing a baseline classifier on first unreviewed transaction features of a first unreviewed transaction record to obtain a first baseline account identifier;
executing the comparison model on (i) a first unreviewed transaction vector of the first unreviewed transaction record and (ii) a plurality of reviewed transaction vectors to obtain a first plurality of comparison scores, the plurality of reviewed transaction vectors corresponding to a plurality of reviewed transaction records each having a user-approved account identifier;
selecting, using the first plurality of comparison scores, a first reviewed transaction record of the plurality of reviewed transaction records, the first reviewed transaction record corresponding to a first comparison score of the first plurality of comparison scores, the first comparison score corresponding to a first user-approved account identifier of the first reviewed transaction record;
selecting, using the first comparison score, one of the first baseline account identifier and the first user-approved account identifier to obtain a first selected account identifier; and presenting the first selected account identifier for the first unreviewed transaction record.
|