Transformer Entity Recognition for Noisy Transaction Names
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing transaction servers face challenges in accurately identifying named entities in transaction records due to noisy data, misspellings, abbreviations, and non-normalized names, leading to false positives and negatives, which can impact transaction authentication and approval processes.
Innovation Solution
Employing a transformer model trained on pairs of transactions to classify non-normalized named entity names to normalized names, using embeddings and similarity analysis to improve accuracy in named entity recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If string similarity metrics are used to identify named entities, then the process is simple and fast, but accuracy deteriorates due to false positives from extraneous information and non-normalized names
Solution Approach 1:
The patent changes the parameter for name comparison from string similarity metrics to normalized string equality. By transforming names to a standardized format (normalization) before comparison, the system achieves accurate identification despite variations in input formats, abbreviations, or misspellings in the original data.
Solution Approach 2:
The patent replaces the mechanical string similarity calculation approach with a transformer-based neural network model. This substitution enables the system to understand semantic relationships and contextual meaning, allowing accurate named entity recognition even when string similarity is low due to normalization differences.
2Reliability
If traditional named entity recognition methods are used, then processing is fast, but reliability deteriorates due to false positives and negatives impacting transaction authentication
Solution Approach 1:
The system performs preliminary name normalization and embedding generation before the actual authentication decision. By pre-processing names to their standardized representations and computing semantic embeddings in advance, the system prepares data structures that enable fast and reliable matching during transaction authentication without sacrificing speed for accuracy.
Solution Approach 2:
The patent introduces transformer model embeddings as an intermediary representation between raw transaction names and authentication decisions. These embeddings serve as a bridge that captures semantic meaning and enables reliable entity matching while maintaining processing efficiency through optimized retrieval and comparison mechanisms.
Data Source
AI summary
A server uses a transformer model to identify a named entity associated with a transaction record. The server receives a transaction record including a text string that includes a non-normalized version of a name of a named entity. The server generates a first embedding of the text string using a first transformer model and identifies a set of similar transactions by comparing the first embedding to second embeddings representing the similar transactions. The server inputs the text string of the transaction record and the set of similar transactions into a second transformer model. The server receives an output from the second transformer and determines that the output indicates that the non-normalized version of the name in the transaction record is classifiable to one of the normalized named entities in the list. The server associates the transaction record with the normalized named entity to which the non-normalized named entity is classifiable.