Transaction Rule Allocation Using Embeddings and Past Transaction Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing transaction monitoring systems face challenges in efficiently managing and processing transaction data due to the use of different data schemas by various third-party software platforms, leading to increased processing and network bandwidth consumption.
Innovation Solution
A computing server utilizes a machine-learned language model to generate embeddings of transaction data and compare them to past transactions, allocating the data to predefined rules based on similarity, thereby reducing processing and network bandwidth consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional transaction monitoring systems process transaction data from multiple third-party platforms with different data schemas, then comprehensive transaction monitoring is achieved, but processing resource consumption and network bandwidth usage increase significantly
Solution Approach 1:
The patent introduces an intermediary embedding layer that transforms transaction data from different third-party platforms into a unified latent space representation. This embedding layer acts as a mediator that standardizes heterogeneous data schemas without requiring direct processing of raw data from each platform, thereby reducing computational overhead while maintaining monitoring accuracy.
Solution Approach 2:
The system transforms transaction data from its original heterogeneous format into a standardized embedding representation in latent space. This parameter transformation converts diverse data schemas into a common mathematical representation, enabling efficient comparison and processing while preserving the essential characteristics needed for fraud detection and monitoring.
2Measurement precision
If traditional systems process and compare raw transaction data from multiple sources, then accurate transaction allocation is achieved, but network bandwidth consumption increases
Solution Approach 1:
The patent extracts only the essential features of transaction data by converting them into compact embedding representations. Instead of transmitting and processing complete raw transaction records, the system extracts key characteristics into condensed vector forms, significantly reducing network bandwidth requirements while retaining the information necessary for accurate transaction allocation and comparison.
3Adaptability or versatility
If manual transaction allocation methods are used across multiple user accounts with different rules, then flexibility in rule application is maintained, but processing time and computational resources increase
Solution Approach 1:
The embedding layer provides a universal representation framework that works across multiple user accounts and transaction rules simultaneously. By transforming diverse transaction data into a common latent space, the system enables a single comparison mechanism to handle multiple users with different rules, eliminating the need for separate processing pipelines for each user while maintaining individual rule flexibility.
Data Source
AI summary
A system for annotating transactions includes a computing server configured to receive transaction data of a transaction associated with a user account of a user. The user is associated with one or more transaction rules. The computing server generates a first embedding from the transaction data, the first embedding being in a latent space of a first machine-learned model. The computing server identifies a set of past transactions of the user. The server compiles a prompt that includes the transaction and the set of past transactions of the user and inputs the prompt to a machine-learned language model to request the machine-learned language model to allocate the transaction into one of the transaction rules. The server receives an output from the machine-learned language model and allocates the transaction data to the one of the transaction rules based on the output from the machine-learned language model.