Neural Embeddings for Sparse Transaction Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning models struggle to effectively learn and analyze high-dimensional, sparse transaction networks due to their inability to scale and retain relevant information about relationships between entities, leading to loss of data in dimensionality reduction techniques.
Innovation Solution
A neural network is trained using a network graph of transaction data, with an embeddings layer that generates low-dimensional dense representations of entities, allowing for the retention of all data and capturing relationships between entities, even in high-dimensional and sparse networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional machine learning models use dimensionality reduction techniques, then the data becomes manageable in size, but significant amounts of relevant information about relationships between entities is lost
Solution Approach 1:
The patent transforms the high-dimensional sparse transaction network into a low-dimensional dense embedding space while preserving relationship information. Neural network embeddings map entities into a compressed vector space where relationships are encoded in the vector representations, achieving dimensionality reduction without information loss through learned feature transformations.
Solution Approach 2:
The patent changes the parameter representation from high-dimensional sparse indicators to low-dimensional dense vectors. By transforming the data representation parameters through neural network embeddings, the system achieves manageable data size while maintaining the essential relationship characteristics through optimized vector representations.
2Device complexity
If conventional machine learning models reduce sparsity to a manageable size, then processing becomes feasible, but relevant information on entity relationships is lost
Solution Approach 1:
The patent applies dimensionality transformation by embedding entities from a sparse high-dimensional space into a dense low-dimensional space. This transformation makes processing feasible while preserving relationship information through the learned embedding vectors that capture entity interactions in the compressed representation.
Solution Approach 2:
The patent replaces conventional mechanical dimensionality reduction techniques with neural network-based embedding learning. Instead of using traditional methods that lose information, the system uses learned neural representations that adaptively preserve relationship structures through training on the transaction data.
3Loss of information
If the full dimensions of the transaction network are learned, then complete relationship information is retained, but conventional models cannot scale to learn weights for all dimensions
Solution Approach 1:
The patent resolves the scaling problem by transforming the learning task from high-dimensional weight estimation to low-dimensional embedding learning. Neural networks efficiently learn compact vector representations for entities, enabling the model to scale to large transaction networks while retaining relationship information through the dense embedding space.
Solution Approach 2:
The patent changes the learning parameters from high-dimensional relationship weights to low-dimensional entity embeddings. This parameter transformation enables scalable learning because the number of embedding parameters grows linearly with the number of entities, whereas learning full relationship matrices would grow quadratically, making the system scalable to large networks.
Data Source
AI summary
Systems, methods, and computer program products to provide neural embeddings of transaction data. A network graph of transaction data based on a plurality of transactions may be received. The network graph of transaction data may define relationships between the transactions, each transaction associated with at least a merchant and an account. A neural network may be trained based on training data comprising a plurality of positive entity pairs and a plurality of negative entity pairs. An embedding function may then encode transaction data for a first new transaction. An embeddings layer of the neural network may determine a vector for the first new transaction based on the encoded transaction data for the first new transaction. A similarity between the vectors for the transactions may be determined. The first new transaction may be determined to be related to the second transaction based on the similarity.