Heterogeneous Graph Prediction Network for Cold Start Conversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing prediction networks struggle to accurately predict a conversion index for users or resources with no recorded conversion behavior, leading to ineffective feature transmission and prediction accuracy issues.
Innovation Solution
A data processing method involving the construction of heterogeneous and homogeneous graphs to train prediction networks, where object and resource features are propagated across graphs to learn conversion behaviors, even for isolated nodes, thereby improving prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a prediction network is trained using existing conversion behavior data, then prediction accuracy is improved for users with conversion history, but prediction accuracy deteriorates for users or resources with no conversion behavior
Solution Approach 1:
The patent applies preliminary action by constructing homogeneous graphs for objects and resources in advance, organizing their features before prediction is needed. This pre-structuring of data enables the prediction network to handle cold start scenarios effectively, as the feature relationships are already established and ready for propagation during prediction
Solution Approach 2:
The patent introduces homogeneous graphs as intermediary structures that mediate between raw feature data and the prediction network. These graphs serve as a bridge that enables feature propagation from objects to resources, allowing the system to generate predictions even when direct conversion behavior data is unavailable
2Loss of information
If the prediction network relies on conversion behavior data for training, then feature transmission is effective for existing users, but feature transmission becomes ineffective for users or resources with no conversion behavior
Solution Approach 1:
The patent transitions from a single-dimension approach (direct conversion behavior data) to a multi-dimensional approach by introducing homogeneous graphs that capture feature relationships in additional dimensions. This allows feature information to be transmitted through graph structures rather than relying solely on direct conversion history
3Productivity
If traditional prediction methods are used, then the system works well for established users, but the system cannot accurately predict for new users or resources (cold start problem)
Solution Approach 1:
The patent applies universality by designing a prediction network that can handle both warm start and cold start scenarios through the same homogeneous graph structure. The system universally processes all prediction requests whether conversion behavior data exists or not, using feature propagation through the graph to generate predictions in both cases
Data Source
AI summary
This application discloses a data processing method performed by a computer device, and the method includes: obtaining a heterogeneous conversion graph including N object nodes and M resource nodes, when an object has a conversion behavior for a resource, a connecting edge exists between a corresponding object node and a corresponding resource node; obtaining a homogeneous object graph corresponding to each object, the graph including object feature nodes of the corresponding object in a plurality of dimensions; obtaining a homogeneous resource graph corresponding to each resource, the graph including resource feature nodes of the corresponding resource in a plurality of dimensions; and training a prediction network based on the heterogeneous conversion graph, the homogeneous object graph of each object, and the homogeneous resource graph of each resource, to obtain a trained prediction network configured to predict a conversion index of an object of interest for a resource of interest.