Character-Level Neural Networks for Transaction Data Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional data enrichment techniques for transaction data, such as text mining and natural language featurization, face challenges in handling varied formats and missing characters, leading to insufficient results in categorization and entity recognition, especially for service providers.
Innovation Solution
The use of character-level deep neural networks, specifically one-dimensional convolutional neural networks (CNN) and recurrent neural networks (RNN) with long short-term memory (LSTM) models, for processing transaction records to classify and tag entities, enabling accurate identification of service providers and other entities within transaction data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional text mining and natural language featurization techniques are used, then processing is simpler, but categorization accuracy and entity recognition performance are insufficient
Solution Approach 1:
The patent segments the transaction data processing into character-level featurization and word-level classification stages. The character-level CNN extracts features from individual characters, which are then aggregated to form word representations for classification. This segmentation allows the system to handle varied formats and missing characters effectively while maintaining manageable model complexity through modular architecture.
Solution Approach 2:
The patent introduces character-level features as an intermediary between raw transaction data and final classification. Instead of directly classifying raw text with complex models, the system first transforms characters into numerical features through CNN, which then serve as input for classification. This intermediary representation layer simplifies the overall processing while improving accuracy.
2Adaptability or versatility
If multiple classifiers are used to handle different service providers, then coverage improves, but data collection and maintenance complexity increase
Solution Approach 1:
The patent implements a universal character-level CNN model that can handle multiple service providers and transaction formats simultaneously. Instead of maintaining separate classifiers for different service providers, the single model processes all transaction data by learning character-level patterns, achieving high coverage across diverse providers while simplifying maintenance to a single model.
Solution Approach 2:
The patent changes the fundamental parameters of the classification approach by moving from word-based or provider-specific features to character-level features. This parameter change enables the model to adapt to various service providers and transaction formats without requiring separate classifiers, as the character-level representation captures universal patterns across all providers.
3Measurement precision
If token-based models are used for entity recognition, then processing is faster, but accuracy in identifying service providers and entities is reduced
Solution Approach 1:
The patent transitions from token-based (word-level) processing to character-level processing, adding a finer granularity dimension to the analysis. By processing data at the character level rather than word level, the system captures subtle patterns in service provider names and entities that token-based models miss, improving recognition accuracy while the convolutional architecture maintains efficient processing.
Data Source
AI summary
Methods, systems and computer program products implementing character-level deep neural networks for information extraction are disclosed. A system uses character-level information retrieved from a transaction record to classify the transaction as a whole and to tag individual sections of the transaction record by entity type. The system processes the transaction record using multiple and separate character-level models. The system can use a one-dimensional neural network for featurization fed into a fully connected network for classification for identifying the most common classes of a transaction record. The system can identify one or more entities, e.g., service provider names, from the transaction using an RNN. The RNN can include one or more LSTM models. The LSTM models can be BI-LSTM models.