Positive-Unlabeled Link Prediction for Edge-Incomplete Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional link prediction models for edge-incomplete graphs fail to consider unobserved edges, leading to overfitting and inability to propagate information between unconnected nodes, thus requiring a new approach that accounts for unconnected node pairs.
Innovation Solution
A link prediction model using positive-unlabeled data learning, where observed edges are treated as positive data and unconnected node pairs as unlabeled data, with random variables representing connection states, trained through a dual loss function and correction loss to balance edge connections and prevent overfitting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional link prediction models are trained using only observed edges as positive data, then the model can be trained on available data, but the model overfits to the given edge-incomplete graph and cannot propagate information between unconnected nodes
Solution Approach 1:
The patent applies preliminary action by introducing random variables to represent connection states of unconnected node pairs before training. This allows the model to pre-establish a framework for handling unobserved edges, enabling information propagation between unconnected nodes during training without requiring actual observed data for these pairs.
Solution Approach 2:
The patent uses random variables as intermediaries to represent the connection states of unconnected node pairs. These random variables act as mediators between the observed edges and the unconnected nodes, allowing the model to propagate information through unconnected nodes by treating their connection states as probabilistic rather than fixed.
2Adaptability or versatility
If the model treats unconnected node pairs as unlabeled data, then the model can consider unobserved edges during training, but the training process becomes more complex requiring dual loss function and correction loss
Solution Approach 1:
The patent segments the training process into distinct components: a dual loss function for handling positive and unlabeled data, and a correction loss function for preventing overfitting. This segmentation allows the complex task of training on edge-incomplete graphs with unobserved edges to be broken down into manageable parts, each addressing specific challenges.
Solution Approach 2:
The patent changes the training parameters by introducing a dual loss function that combines different loss components with specific weights. This allows the model to balance the influence of observed edges and unconnected node pairs during training, managing complexity through parameter adjustment rather than structural simplification.
3Reliability
If the model propagates information through unconnected nodes, then link prediction accuracy improves, but computational complexity increases
Solution Approach 1:
The patent substitutes the mechanical system of traditional graph traversal with a probabilistic approach using random variables. Instead of mechanically traversing all possible paths through unconnected nodes, the model uses probabilistic representations to capture information flow, reducing computational complexity while maintaining accuracy.
Data Source
AI summary
Proposed herein are a link prediction method and apparatus. The link prediction method that is performed by the link prediction apparatus includes predicting one or more edges having a probability of being connected in the structure of an edge-incomplete graph by entering the edge-incomplete graph into a link prediction model. The link prediction model is a model that performs binary classification by processing at least one edge observed in the structure of the edge-incomplete graph as positive data and processing at least one node pair unconnected in the structure of the edge-incomplete graph as unlabeled data.