Entity Transaction Attribute Prediction via Temporal Graph Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI-based systems struggle to accurately predict entity transaction attributes due to the influence of unexpected events, leading to reduced prediction accuracy.
Innovation Solution
An entity transaction attribute determination method and apparatus that utilizes historical relational networks to extract node vectors based on descriptive information, performs vector fusion of neighbor nodes, and processes these vectors through a pre-trained time-series neural network to determine attribute states or categories, effectively capturing time-series characteristics and environmental changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional AI-based prediction methods are used to predict entity transaction attributes, then the prediction process is simple and fast, but the prediction accuracy deteriorates due to the inability to capture time-series characteristics and unexpected events
Solution Approach 1:
The patent segments the prediction task into multiple time points, constructing historical relational networks at different time points (t-1, t-2, ..., t-n) to capture temporal evolution. Each time point's relational network is processed separately through graph neural networks, and the results are integrated through time-series neural networks, enabling the model to capture both local temporal patterns and global time-series characteristics.
Solution Approach 2:
The patent adds a time dimension to the traditional relational network by constructing historical relational networks at multiple time points. This transforms the static prediction problem into a dynamic time-series prediction problem, allowing the model to capture unexpected events and temporal evolution by comparing changes across different time dimensions.
2Measurement precision
If the model considers more factors including unexpected events and time-series characteristics, then the prediction accuracy improves, but the computational complexity increases
Solution Approach 1:
The patent performs preliminary actions by pre-processing descriptive information into node vectors and pre-training the time-series neural network on historical data. The graph neural networks are also pre-trained to capture structural patterns. This preliminary processing reduces the computational burden during actual prediction, as the model has already learned temporal patterns and structural relationships in advance.
Solution Approach 2:
The model uses self-service mechanisms through automated vector fusion and time-series processing. The graph neural networks automatically aggregate neighbor node vectors, and the time-series neural networks automatically capture temporal dependencies, reducing the need for manual feature engineering and lowering overall computational complexity.
3Adaptability or versatility
If historical relational networks with multiple time points are constructed to capture temporal evolution, then the ability to handle unexpected events improves, but the data processing complexity increases
Solution Approach 1:
The patent introduces vector fusion as an intermediary mechanism between the historical relational networks and the final prediction. The graph neural networks serve as intermediaries that aggregate information from multiple sources (entity's own attributes, neighbor nodes, temporal relationships) into unified node vectors, which are then processed by time-series neural networks. This intermediary processing simplifies the complexity of directly handling multi-time-point relational data.
Data Source
AI summary
An entity transaction attribute determination method for determining an attribute state or an attribute category of a to-be-predicted entity on a preset transaction is provided. The method comprises obtaining a plurality of historical relational networks sequentially arranged under a temporal order; determining, for each of the historical relational networks and through vector fusion of neighbor nodes, a plurality of description vectors of the to-be-predicted entity; processing, through a pre-trained time-series neural network, the description vectors to obtain an output result; and determining, according to the output result, the attribute state or the attribute category of the preset transaction attribute for the to-be-predicted entity. The method improves the accuracy of predicting a preset transaction attribute of an entity through the analysis of the description vectors.