Graph Filter Rating Prediction Reducing Computational Overhead
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recommender systems consume large computing resources and fail to effectively utilize all available information for user rating predictions, leading to suboptimal recommendations.
Innovation Solution
A system that uses graph filters to compute user rating predictions by applying band-stop filters to user-item rating data, considering similarity networks between users and items, allowing for efficient interaction and high-order interactions without significant increases in computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing recommender systems use traditional collaborative filtering methods, then they can provide recommendations based on user behavior data, but they consume large amounts of computing resources
Solution Approach 1:
The patent segments the rating prediction process by applying graph filters to decompose the user-item rating matrix into low-rank components. This segmentation allows the system to process and predict ratings by analyzing smaller, filtered components rather than the complete rating matrix, thereby reducing computational resource consumption while maintaining prediction accuracy.
Solution Approach 2:
The patent changes the parameter representation by transforming the rating data into the graph frequency domain using graph filters. By operating in this transformed domain and applying spectral filtering, the system can achieve accurate predictions with reduced computational complexity compared to traditional methods that operate directly in the original rating space.
2Use of energy by moving object
If existing recommender systems use simple filtering methods, then they consume fewer computing resources, but they fail to account for high-order interactions between users and items
Solution Approach 1:
The patent introduces a new dimension by transforming the rating data into the graph frequency domain. This dimensional transformation enables the system to capture high-order interactions between users and items through spectral analysis, allowing simple filtering operations in the frequency domain to implicitly model complex interaction patterns that would require much more complex operations in the original domain.
Data Source
AI summary
Methods, systems, and computer readable media for computing user rating predictions are disclosed. In some examples, a recommender is implemented on at least one processor and memory and programmed for receiving a request to select one or more recommended items for a user from unrated items that are unrated by the user to the recommender. The recommender is programmed for computing user rating predictions for each of the unrated items by applying a graph filter to user-item rating data. The user-item rating data includes other-user ratings of the unrated items by other users and a same-user ratings of other items by the user. The recommender is programmed for selecting the one or more recommended items from the unrated items using the user rating predictions for the unrated items and outputting the recommended items in response to the request.