Recommendation System Sampling for Reduced Calculation Load
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current website recommendation systems face high calculation loads when recommending potential users to network service providers using collaborative filtering algorithms, which affects recommendation efficiency.
Innovation Solution
A method and apparatus that identify core users and sample users based on network-related behaviors, calculate similarity between sample users, and recommend potential users to network service providers, reducing calculation load through sampling and similarity processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If collaborative filtering algorithm is used to recommend potential users to network service providers, then recommendation accuracy is improved, but calculation load increases heavily
Solution Approach 1:
The patent segments the large user set into multiple smaller subsets based on user behaviors and preferences. By dividing the user base into manageable segments, the system can perform collaborative filtering calculations on smaller groups, reducing overall calculation load while maintaining recommendation accuracy through targeted segmentation based on user characteristics
Solution Approach 2:
The patent performs preliminary user profiling and behavior analysis before executing the main recommendation algorithm. User data is pre-processed to create behavioral profiles, preferences, and segmentation categories in advance, so that when the collaborative filtering algorithm runs, it operates on pre-organized data structures rather than raw user data, significantly reducing real-time calculation requirements
2Quantity of substance
If collaborative filtering algorithm processes all users, then comprehensive user coverage is achieved, but recommendation efficiency decreases
Solution Approach 1:
The system segments users into different groups based on behavior patterns, preferences, and engagement levels. This segmentation allows the recommendation engine to process users in manageable batches rather than as a monolithic set, improving processing efficiency while ensuring all user segments are covered through systematic iteration through each group
Solution Approach 2:
The patent implements a multi-pass recommendation approach where not all users are processed in every recommendation cycle. Instead, the system processes representative samples from each user segment across multiple passes, achieving comprehensive coverage over time without the computational overhead of processing every single user in each cycle
Data Source
AI summary
Disclosed is a recommendation method and apparatus. The recommendation method comprises: identifying at least one core user of a network service provider and a set of users who have demonstrated one or more network-related behaviors with respect to a first predetermined association item; sampling the set of users to identify a set of sample users; calculating a similarity between the sample users to identify a list of similar users; and sending a recommendation of potential users to a network service provider, the recommendation of potential users based on the list of similar users and the at least one core user of a network service provider. Potential users may be efficiently recommended to a network service provider.