Transaction Classification via Industry Name Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning techniques for predicting transaction classifications are limited by their reliance on limited data points, leading to inaccurate predictions.
Innovation Solution
A method involving the use of industry name embeddings and transaction embeddings to train a machine learning model, allowing it to predict transaction classifications more accurately by considering features related to the payor's industry.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing machine learning techniques use limited training data points (monetary amount and payee description), then the training process is simple and fast, but the prediction accuracy is limited
Solution Approach 1:
The patent applies preliminary action by generating embeddings for payors and payees in advance before the actual transaction classification task. These embeddings capture semantic information from historical transactions and are stored for later use, allowing the model to leverage pre-processed semantic representations rather than processing raw text during classification, thus improving accuracy while managing complexity
Solution Approach 2:
The patent introduces embeddings as an intermediary representation between raw transaction data and classification predictions. The embedding models transform payor and payee names into dense vector representations that capture semantic meaning, serving as a bridge that enables more accurate classification without directly processing complex unstructured text during the main classification process
2Measurement precision
If more training data and features are incorporated to improve prediction accuracy, then classification precision improves, but computing resource utilization increases
Solution Approach 1:
The embedding models are trained separately in advance on historical transaction data, performing the computationally intensive feature extraction beforehand. During actual transaction classification, the pre-trained embeddings are reused, significantly reducing computing resource requirements for real-time predictions while maintaining high classification precision
Solution Approach 2:
The patent transforms categorical text data (payor and payee names) into continuous embedding vectors, changing the parameter representation from discrete text to continuous numerical space. This transformation enables more efficient computation during classification while capturing richer semantic information, improving precision without proportionally increasing computational cost
Data Source
AI summary
A method for training a machine learning model to automatically predict a classification for an uncategorized transaction includes: retrieving a plurality of historical transactions involving a plurality of different payors and a plurality of different payees; generating a first plurality of embeddings based, at least in part, on a subset of the historical transactions having a first type, wherein each of the first plurality of embeddings is representative of an industry name associated with one or more payors involved in the first set of historical transactions; generating training data for the machine learning model based, at least in part, on the first plurality of embeddings; and training the machine learning model through a supervised learning process using the training data.