Persona-Augmented Graph Learning for Interpretable Purchase Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recommendation systems using graph neural networks struggle to learn the relative relationship between users and personas, leading to poor interpretability and difficulty in determining which persona is likely to purchase an item.
Innovation Solution
Generate additional graph data by adding personas as nodes, extract subject and persona representations, and calculate recommendation and comparison losses to update the model, separating user and persona representations to improve interpretability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If graph neural network is used to learn purchase probability between users and items, then recommendation accuracy is improved, but interpretability deteriorates because relative relationship between user and persona cannot be learned
Solution Approach 1:
The patent segments the user representation into two distinct components: user representation and persona representation. This segmentation allows the model to separately learn and interpret the contributions of different user aspects to purchase decisions, thereby maintaining recommendation accuracy while improving interpretability. The loss function is also segmented into multiple components (recommendation loss, persona discrimination loss, user-persona discrimination loss) that independently optimize different aspects of the model.
Solution Approach 2:
The patent introduces persona as an intermediary representation that mediates between the user and items. The persona nodes serve as intermediaries that capture specific preference aspects, allowing the model to explain recommendations through persona-item relationships while maintaining accurate user-item purchase probability predictions. This intermediary structure enables interpretability without sacrificing recommendation performance.
2Productivity
If relative levels of purchase probability are learned, then recommendation performance is improved, but the ability to determine which user or persona is likely to purchase an item deteriorates
Solution Approach 1:
The patent segments the purchase probability learning into distinct components: overall user-item purchase probability and persona-specific purchase probability. This segmentation enables the model to maintain high recommendation performance while providing clear, interpretable information about which specific persona is likely to purchase each item, thus improving ease of determination.
Solution Approach 2:
The patent adds a new dimension to the representation space by introducing persona nodes as a separate dimensional layer. This dimensional expansion allows the model to capture both aggregate user preferences and specific persona preferences simultaneously, enabling high recommendation performance while providing clear interpretability about which persona drives each recommendation.
3Loss of information
If personas are added to the graph data, then interpretability is improved, but model complexity increases
Solution Approach 1:
The patent segments the graph structure into distinct components: original user-item edges and new persona-item edges. This segmentation allows the model to incorporate persona information for improved interpretability while maintaining a structured, manageable graph architecture. The loss function is also segmented into separate components that independently optimize different aspects, preventing complexity from becoming unmanageable.
Solution Approach 2:
The patent designs the persona nodes to serve multiple functions simultaneously: they capture user preferences for interpretability, serve as intermediaries for recommendation generation, and provide discriminative features for distinguishing different user aspects. This multi-functionality reduces the need for additional separate mechanisms, thereby limiting the increase in model complexity despite the added interpretability capability.
Data Source
AI summary
According to one embodiment, an information processing apparatus includes a processor. The processor acquires graph data in which nodes of subjects and edges of a relationship between the subjects. The processor generates additional graph data in which one or more personas are added to the graph data as the nodes. The processor extracts subject representations and persona representations from the additional graph data, using a model for extracting representations. The processor calculates first loss depending on difference between similarity between nodes connected by the edge and similarity between nodes not connected by the edge. The processor calculates second loss depending on difference between similarity between node of the subject and node connected to the node of the subject, and similarity between node of the persona and node connected to the node of the persona. The processor updates the model, using the first and second losses.