Transaction Account Matching via Embedding Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional financial management systems are unable to accurately and efficiently assist users in sorting financial transactions into the proper accounts, leading to wasted time and resources, and potential user dissatisfaction due to manual sorting processes and incorrect account recommendations.
Innovation Solution
The system generates account grouping data, transaction grouping data, and user grouping data to analyze and predict the most suitable account for new financial transactions based on past user categorizations, merchant involvement, and transaction characteristics, providing recommendations or automatic sorting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional financial management systems use account names for matching transactions, then the system structure remains simple, but the accuracy of account recommendations deteriorates
Solution Approach 1:
The system transforms account names into numerical feature vectors that capture semantic meaning and usage patterns. By representing accounts in a numerical space, the system can perform mathematical operations to find similar accounts, dramatically improving recommendation accuracy while maintaining manageable complexity through vector space modeling
Solution Approach 2:
The patent introduces an intermediary embedding model that translates account names into numerical representations. This intermediary layer enables accurate matching by converting textual account names into a format that allows for similarity computation, bridging the gap between simple name-based systems and complex recommendation needs
2Loss of time
If users manually sort transactions through the entire chart of accounts, then account recommendation accuracy is not compromised, but the time required for sorting deteriorates
Solution Approach 1:
The system pre-computes account embeddings and organizes accounts into hierarchical groups based on their numerical representations before transactions need to be sorted. This preliminary structuring allows for rapid matching when transactions arrive, eliminating the need for users to manually browse the entire chart of accounts and dramatically reducing sorting time
Solution Approach 2:
The patent replaces the mechanical manual sorting process with an automated computational system. Instead of users physically navigating through account lists, the system uses numerical matching and similarity computations to automatically identify and recommend appropriate accounts, substituting human manual effort with algorithmic processing
3Productivity
If traditional systems provide no account recommendations, then system complexity remains low, but productivity deteriorates due to manual sorting
Solution Approach 1:
The system enables transactions to self-categorize by automatically matching them with appropriate accounts based on numerical similarity. The embedding model and matching algorithm work autonomously to recommend and assign accounts without requiring manual user intervention, significantly boosting productivity while implementing practical automation
Solution Approach 2:
The system incorporates feedback mechanisms where user corrections to automated account recommendations are used to refine and improve the embedding models over time. This feedback loop allows the system to learn from user behavior and continuously enhance its automation capabilities, increasing productivity while adapting to user needs
4Adaptability or versatility
If the system uses user-specific transaction history for account matching, then personalization improves, but the system cannot handle new merchants or transactions without prior user history
Solution Approach 1:
The patent creates a universal account embedding space that captures general accounting patterns and relationships applicable across all users. This universal representation allows the system to handle new transactions and merchants by comparing them against the general account structure, enabling the system to function effectively even when specific user historical data is unavailable
Solution Approach 2:
The system merges user-specific transaction patterns with general accounting knowledge represented in the embedding model. By combining individual user behavior with universal account relationships, the system can provide personalized recommendations while also handling new merchants and transactions that lack user-specific history, leveraging both specific and general information
Data Source
AI summary
A method and system groups user accounts in a financial management system based on the similarities of the financial transactions associated with the accounts. The method and system groups merchants involved in the financial transactions based on how financial transactions involving the merchants are sorted into groups of merchants. The method and system group users based on how the users sort financial transactions into their accounts. The method and system assist users to sort future financial transactions based on the groups of accounts, the groups of merchants, and the groups of users.