Neural Collaborative Filtering for Payment Offer Recommendations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional neural network systems for AI applications, such as recommender systems, face computational burdens due to the need for feature engineering and staging steps, especially when dealing with updated data frequently, which can be prohibitive in certain use cases.
Innovation Solution
A deep learning system utilizing neural collaborative filtering (NCF) and wide and deep learning techniques is applied to payment card transaction data to generate models that predict interactions between consumers and merchants, reducing computational efforts by automatically performing internal feature selection and optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional neural network systems are used with feature engineering and staging steps, then predictive modeling capability is achieved, but computational burden increases significantly
Solution Approach 1:
The patent extracts and removes the computationally expensive feature engineering and staging steps from the traditional neural network pipeline. By using raw transaction data directly as input to the neural collaborative filtering model, the system eliminates the need for separate feature extraction modules, thereby reducing computational burden while maintaining predictive accuracy through the model's ability to automatically learn relevant features from raw data.
Solution Approach 2:
The neural collaborative filtering model performs self-service by automatically performing internal feature selection and optimization without requiring external feature engineering. The model learns latent representations and interactions directly from the input data, making the system self-sufficient and eliminating the need for manual feature preparation, thus reducing computational overhead.
2Measurement precision
If feature engineering is performed regularly to handle updated data, then model accuracy is maintained, but processing time increases
Solution Approach 1:
The system enables continuous learning and model updating without interrupting the data pipeline for feature engineering. The neural collaborative filtering model can be retrained incrementally on updated transaction data, maintaining model accuracy while avoiding the time-consuming feature engineering process. The model continuously adapts to new data patterns through its architecture that directly processes raw transaction data.
3Measurement precision
If comprehensive feature engineering is applied to capture indirect relationships, then predictive accuracy improves, but system complexity increases
Solution Approach 1:
The patent merges the feature engineering function into the neural collaborative filtering model itself. Instead of having separate modules for feature extraction, selection, and transformation, the system combines these functions into the model's latent factor decomposition process. The model automatically captures indirect relationships through matrix factorization and neural network layers, eliminating the need for complex external feature engineering pipelines.
Data Source
AI summary
A system for training and applying deep learning to generate offer recommendations within a payment card network includes a propensity engine configured to train a machine learning model using neural collaborative filtering. The model is trained based on payment card transaction data providing indications of cardholder interactions with merchants, the model is configured to generate offer recommendations for cardholders of payment cards made available by merchants. The system also includes a model serving engine configured to apply available offer data to the model, thereby generating a set of offer recommendations matching one or more available offers with one or more cardholders. The system also includes an execution and fulfilment engine configured to deliver an available offer to a cardholder and fulfill the available offer when activated by the cardholder.