Graph Neural Network Vector Update for Interaction Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional interaction prediction systems are resource-intensive and inefficient, particularly when dealing with large volumes of interactions, leading to unnecessary processing power usage and reduced approval rates for resource providers due to repeated retries of declined interactions.
Innovation Solution
A machine learning-based system using a graph neural network framework with recurrent neural networks to predict future interactions and their approval probabilities, updating vector representations dynamically based on new interaction data, and aggregating information from neighboring nodes to improve prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional interaction prediction systems re-create or re-format entire prediction models as new data is obtained, then prediction accuracy can be maintained, but computing resources and processing time are significantly consumed
Solution Approach 1:
The patent segments the prediction model into static components (model architecture, feature definitions) and dynamic components (interaction data, vector representations). Only the dynamic components are updated incrementally with new data, while static components remain unchanged. This segmentation allows the system to maintain prediction accuracy without re-processing the entire model structure, significantly reducing computing resource consumption.
Solution Approach 2:
The system performs preliminary actions by pre-defining the model architecture, feature extraction pipelines, and vector representation schemas during an initial setup phase. These preliminary configurations remain static and do not need to be re-created when new interaction data arrives, allowing the system to focus computational resources only on updating the necessary dynamic elements.
2Reliability
If resource providers retry declined interactions multiple times, then approval rate may improve, but processing power is wasted and overall approval rate decreases
Solution Approach 1:
The system implements feedback mechanisms by analyzing the outcomes of previous interaction attempts and using this information to predict the likelihood of future approvals. The prediction model provides feedback to resource providers about which interactions are worth retrying and which should be abandoned, enabling intelligent retry strategies that improve approval rates while minimizing wasted processing power on doomed attempts.
Solution Approach 2:
Instead of uniformly retrying all declined interactions or abandoning all of them, the system applies partial action by selectively retrying only those interactions that the prediction model identifies as having a high probability of approval. This selective approach avoids the excessive processing power consumption of blanket retry strategies while still capturing the benefits of retrying promising interactions.
3Measurement precision
If vector representations are updated for each new interaction, then prediction accuracy improves, but system complexity and processing overhead increase
Solution Approach 1:
The system applies dynamics by making vector representations mutable and adaptable rather than static. Vector representations are updated incrementally as new interaction data becomes available, allowing the model to dynamically adapt to changing patterns in the data. This dynamic updating mechanism maintains high prediction accuracy without requiring complete model re-creation, balancing accuracy improvements with manageable system complexity.
Data Source
AI summary
A method includes extracting, by an analysis computer, a dataset including initial vector representations for each of a plurality of user nodes and for each of a plurality of resource provider nodes. The analysis computer can then generate updated vector representations as new interaction data arrives over time, and use them to perform predictions of future interactions.