Sign-Aware Graph Neural Network for Recommendation Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recommendation systems fail to accurately consider both preferences and non-preferences in user evaluations, leading to loss of information and inaccurate item recommendations, as they primarily focus on positive interactions without accounting for negative evaluations.
Innovation Solution
A sign-aware recommendation apparatus and method using a graph neural network that separates edges into positive and negative edges based on evaluation scores, generating separate graphs for preferences and non-preferences, and embedding user and item nodes into a common virtual space using pre-trained neural networks to accurately recommend items.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If existing network embedding techniques vectorize nodes by considering only positive interactions, then the model complexity remains relatively simple, but information about user non-preference is lost
Solution Approach 1:
The patent segments the interaction data into positive interactions and negative interactions, creating separate embedding models for each type. This segmentation allows the system to preserve both user preferences and non-preferences without mixing them, thereby reducing information loss while managing complexity through modular architecture
Solution Approach 2:
The patent extends the traditional single embedding space into a multi-dimensional space that accommodates both positive and negative interactions. By adding the dimension of interaction polarity, the system can represent user preferences and non-preferences simultaneously, preventing information loss about user attitudes
2Measurement precision
If all evaluation scores are treated as positive evaluations, then the processing logic remains simple, but recommendation accuracy deteriorates due to loss of negative evaluation information
Solution Approach 1:
The patent segments evaluation scores into positive evaluations (indicating user preference) and negative evaluations (indicating user non-preference). This segmentation enables the system to process and weigh different types of feedback separately, improving recommendation accuracy by capturing the full spectrum of user attitudes rather than treating all scores uniformly
Solution Approach 2:
The patent applies different processing logic to different types of evaluation scores based on their local characteristics. Positive evaluations are processed through one pathway while negative evaluations are processed through another, allowing each type to be handled with appropriate complexity and weight, thereby improving overall measurement precision
3Loss of information
If negative evaluation and irrelevance are not separately reflected, then the data structure remains simple, but information about user preferences becomes less accurate
Solution Approach 1:
The patent segments the data structure to distinguish between negative evaluations (active dispreference) and irrelevance (no evaluation). This segmentation prevents the conflation of users who actively dislike an item with users who simply did not evaluate it, thereby preserving accurate user preference information while organizing data into meaningful categories
Data Source
AI summary
A recommendation apparatus separately extracts a positive graph and a negative graph representing the user's preference and non-preference for each item from the bipartite graph representing the relationship between users and items obtained using evaluation data, and vectorizes a plurality of nodes corresponding to users and items based on each of the extracted positive graph and negative graph and embeds them in a common virtual embedding space to recommend items that the user would prefer by considering not only preferences but also non-preferences of similar users for items.