Unified Entity Detection for Transaction Data Structuring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for structuring transactional data, particularly the unstructured description fields, are inefficient and inaccurate as they independently detect various entities within a transaction description, leading to inconsistencies and a lack of standardized formatting.
Innovation Solution
The proposed system analyzes a transaction as a whole to detect entities collectively, using multiple entity-specific extractors with confidence levels and a conflict resolution component that employs a search tree and user-defined expert rules to minimize overlaps and gaps, ensuring high textual coverage and overall confidence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple entity-specific extractors are used to independently detect entities within transaction descriptions, then various entities can be identified, but the process becomes inefficient and inaccurate with overlapping assignments and gaps
Solution Approach 1:
The patent combines multiple independent entity-specific extractors into a unified system that processes transaction descriptions collectively. Instead of having extractors work independently and then merge results, the system integrates them to detect entities in a coordinated manner, resolving overlaps and gaps through a standardized formatting approach that assigns each entity to its most appropriate extractor based on confidence levels and entity type hierarchies.
Solution Approach 2:
The patent creates a universal entity detection system that handles multiple entity types (merchant, location, date, payment method, etc.) through a single integrated processing framework. This multi-functional system can detect and structure various entities within transaction descriptions using a common approach, eliminating the need for separate independent processing pipelines for each entity type.
2Adaptability or versatility
If entities are detected independently within transaction descriptions, then various entity types can be identified, but inconsistencies and lack of standardized formatting occur
Solution Approach 1:
The patent applies parameter changes by introducing confidence levels and entity type classifications as key parameters for resolving detection results. Each extractor assigns confidence levels to its detected entities, and the system uses these parameters along with entity type hierarchies to determine the most appropriate assignment. This parameter-based approach ensures standardized formatting while maintaining versatility across different entity types.
Solution Approach 2:
The patent introduces an intermediary conflict resolution component that mediates between multiple entity-specific extractors. This intermediary receives detection results from all extractors, resolves conflicts through standardized rules and expert-defined criteria, and produces consistent formatted output. The intermediary ensures reliability and consistency while allowing the system to handle diverse entity types.
3Measurement precision
If transaction descriptions are analyzed multiple times to identify various entities independently, then comprehensive entity detection can be achieved, but the process becomes inefficient
Solution Approach 1:
The patent applies preliminary action by performing entity detection across all entity types in a single coordinated pass through the transaction description. Instead of analyzing the description multiple times for different entity types, the system prepares and executes a comprehensive detection process that identifies all entities simultaneously, reducing processing time while maintaining completeness.
Solution Approach 2:
The patent ensures continuity of useful action by maintaining an active detection process that continuously evaluates the transaction description for all entity types without interruption or repeated passes. The unified system processes the description once, continuously identifying and classifying entities of all types, thereby eliminating the time loss associated with multiple separate analysis cycles.
Data Source
AI summary
Systems and methods are provided to structure event description data.