Transaction Categorization Using Hybrid Vector Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated transaction categorization systems face challenges in accurately categorizing transactions for first-time entities, entities with evolving categorization behaviors, and those with customized charts of accounts, due to limited historical data and complex accounting structures.
Innovation Solution
A method and system that utilize a deep learning framework to leverage populational data for new entities, in-session learning for entities with changing behaviors, and a two-stage process involving transaction vector generation and customized chart of accounts classification for established entities with complex structures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If automated transaction categorization is implemented for first-time entities with limited historical data, then the system can provide initial categorization recommendations, but the categorization accuracy is reduced due to insufficient training data
Solution Approach 1:
The patent combines general population transaction patterns with entity-specific customization data to create hybrid categorization recommendations. For first-time entities, the system merges anonymized population-level transaction data with the entity's initial categorization preferences to generate accurate recommendations even when historical entity data is limited.
Solution Approach 2:
The system performs preliminary categorization using general models before entity customization is established. Default categorization rules and population-based patterns are applied in advance to provide immediate automated categorization functionality, which is then refined as the entity accumulates more data and customizations.
2Adaptability or versatility
If the system allows entities to create customized charts of accounts, then the system provides personalized categorization recommendations, but the device complexity increases due to handling unique account structures
Solution Approach 1:
The patent segments the categorization system into distinct components: general models that handle population-level patterns, customization models that learn entity-specific preferences, and recommendation models that synthesize both. This segmentation allows the system to manage complexity by processing different types of data through specialized sub-systems rather than a monolithic approach.
Solution Approach 2:
The system introduces intermediary layers including vector embeddings that translate diverse account structures into a unified representation space, and match models that bridge general patterns with custom accounts. These intermediaries enable the system to handle personalized charts of accounts without directly processing the full complexity of unique account structures.
3Area of stationary object
If the system handles both new entities with limited data and established entities with complex customizations using a single system, then the system achieves universality, but the difficulty of detecting and measuring transaction patterns increases
Solution Approach 1:
The patent implements dynamic model selection where the system automatically adjusts which models to use based on entity characteristics. For new entities, general models dominate; for established entities with customizations, customization models gain weight. This dynamic adaptation allows a single system to handle diverse entity types while optimizing pattern recognition for each case.
Solution Approach 2:
The system changes parameters such as data weighting, model confidence thresholds, and feature importance based on entity maturity and customization level. For entities with limited history, the system increases reliance on population patterns; for established entities, it shifts weight to entity-specific patterns, thereby managing pattern recognition difficulty through parameter adjustment.
Data Source
AI summary
A transaction model of a general model generates a target transaction vector for a target transaction record. The general model also generates account vectors for accounts. A match score is generated between the account vectors and the transaction vector. The general model selects a first account identifier of an account using the match score. The transaction model also generates historical transaction vectors for historical transaction records. Further, a comparison score is generated between the historical transaction vectors and the target transaction vector. A second account identifier of an historical transaction is selected according to the comparison score. One of the first account identifier and the second account identifier is selected as the account identifier for the transaction record, and the transaction record is stored with the account identifier.