Dimensionality Reduction for Transaction Categorization

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Traditional electronic bookkeeping systems fail to accurately and efficiently assist users in sorting financial transactions into the correct accounts due to their inability to understand the nature of user-created accounts, leading to inefficiencies and potential errors, which can deter users from utilizing these systems.

Innovation Solution

The system generates merchant and account characteristics vectors, reduces their dimensionality using techniques like latent semantic indexing or Word2Vec, and embeds them into a compressed vector space to identify the most likely account for a financial transaction based on proximity, thereby improving sorting accuracy and efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If traditional electronic bookkeeping systems use account names to recommend accounts for sorting transactions, then the system can provide automated account recommendations, but the recommendations are often inaccurate or unrelated because the system cannot understand the true nature of user-created accounts

Engineering Contradiction:

Improveautomated account recommendationVSAvoidaccount recommendation accuracy

Core Design Contradiction:

Extent of automationVSMeasurement precision

Solution Approach 1:

The patent introduces co-occurrence data as an intermediary to bridge the gap between account names and account semantics. By analyzing which merchants co-occur in the same accounts across multiple users, the system builds a knowledge base that captures the true nature of accounts without requiring direct interpretation of account names. This intermediary data structure enables accurate recommendations while maintaining automation.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent replaces the mechanical string-matching approach (comparing account names) with a data-driven semantic understanding approach. Instead of relying on literal name comparisons, the system uses statistical patterns from co-occurrence data to understand account semantics, substituting a simplistic mechanical system with a more sophisticated analytical system.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If traditional systems manually sort transactions through a chart of accounts, then users can accurately categorize transactions, but the process becomes tedious and time-consuming especially on mobile devices

Engineering Contradiction:

Improvetransaction categorization accuracyVSAvoidtime to sort transactions

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent enables the system to serve itself by automatically learning account semantics from aggregated user data and then using this learned knowledge to automatically recommend accounts for new transactions. The system performs the categorization task for itself without requiring manual intervention, while still maintaining high accuracy through its learned understanding of account purposes.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent performs preliminary action by pre-computing co-occurrence statistics and building a knowledge base of account semantics before actual transaction categorization is needed. This advance preparation allows the system to quickly provide accurate recommendations when users need to categorize transactions, eliminating the need for time-consuming manual sorting.

Inventive Principle:

Principle #10Preliminary action

3Extent of automation

If the system recommends faulty accounts to users, then automated sorting can be provided, but users experience distraction and extra work in finding the correct account

Engineering Contradiction:

Improveautomated transaction sortingVSAvoiduser effort to find correct account

Core Design Contradiction:

Extent of automationVSEase of operation

Solution Approach 1:

The patent incorporates feedback mechanisms where the system learns from user corrections and interactions. When users manually adjust account recommendations or correct misclassifications, this feedback is used to refine the co-occurrence models and improve future recommendations. This continuous feedback loop ensures that automation improves over time while reducing user effort.

Inventive Principle:

Principle #23Feedback

4Device complexity

If electronic bookkeeping systems cannot understand account nature, then system complexity remains low, but the potential of the system is under-utilized and users may abandon the system

Engineering Contradiction:

Improvesystem complexityVSAvoidsystem potential utilization

Core Design Contradiction:

Device complexityVSAdaptability or versatility

Solution Approach 1:

The patent segments the problem of understanding account semantics into manageable components: collecting co-occurrence data, building frequency matrices, computing similarity metrics, and generating recommendations. This segmentation allows the system to achieve sophisticated account understanding without overwhelming complexity, breaking down the challenging task into discrete, implementable steps.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS12118622B2System and method for dimensionality reduction of vendor co-occurrence observations for improved transaction categorization

Publication Date: 2024.10.15 INTUIT INC

AI summary

A method and system assist users of an electronic bookkeeping system to sort financial transactions into accounts representing bookkeeping categories. The method and system generate vectors representing the accounts and the merchants known to the bookkeeping system. The method and system generate compressed versions of the vectors by compressing the vectors. The method and system assist users to sort the financial transactions by analyzing the compressed vectors.