Service Usage Prediction via Correlation Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting service usage in communication networks are inefficient when dealing with a large number of services, as they require learning separate models for each service and struggle with exponential combinations of service usage, leading to significant processing overhead.
Innovation Solution
The method involves analyzing the historical usage of multiple services to identify correlations, representing these correlations as a joint representation, and using a subset of representative services to predict future usage, thereby reducing the complexity of the prediction task.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If separate models are learned for each service, then prediction accuracy for individual services is maintained, but processing overhead and computational complexity increase significantly
Solution Approach 1:
The patent segments the service set into two categories: representative services (small subset) and non-representative services (remaining services). Separate prediction models are only learned for representative services, while non-representative services are predicted using correlation with representative services. This segmentation reduces the number of models from N (total services) to K (representative services), where K << N, thereby reducing processing overhead while maintaining prediction accuracy.
Solution Approach 2:
The patent creates a universal prediction framework where a small set of representative services serves multiple purposes: (1) they are directly predicted using learned models, and (2) they serve as basis for predicting all other non-representative services through correlation analysis. This multi-functionality allows the system to handle N services using only K models, reducing computational complexity while maintaining comprehensive prediction capability.
2Measurement precision
If joint representation of all service combinations is used, then prediction accuracy improves by utilizing usage history of all services, but the complexity increases exponentially with the number of services
Solution Approach 1:
The patent extracts a small subset of representative services from the complete service set that capture the essential usage patterns and correlations. Instead of using all N services in the joint representation, only K representative services are selected, where K << N. This extraction reduces the joint representation complexity from 2^N to 2^K, making the problem computationally tractable while preserving the most important predictive information.
Solution Approach 2:
The patent applies partial action by using only a subset of services (representative services) rather than all services for direct model learning. The remaining non-representative services are handled through correlation-based prediction. This partial approach to model learning reduces computational complexity from exponential in N to exponential in K, while still achieving accurate predictions for all services through the combination of direct and correlation-based prediction.
3Measurement precision
If correlation analysis is performed between all service pairs, then prediction accuracy for service usage patterns improves, but computational requirements and time increase
Solution Approach 1:
The patent segments the correlation analysis into two parts: (1) correlation between representative services and non-representative services, and (2) correlation between pairs of non-representative services. By focusing correlation analysis on the representative services as anchors, the system avoids performing exhaustive correlation analysis on all N services, reducing computational time from O(N^2) to approximately O(K*N), where K << N.
Data Source
AI summary
A method in a communication network for predicting whether a second service will be used by a user in a future time slot. The method comprises obtaining information about which services, of a plurality of services, have historically been used in which time slots in the communication network. The method comprises analysing the obtained information to determine that, whether a first service of the plurality of services is used in any of the historic time slots, correlates with whether the second service of the plurality of services is used in said historic time slot. The method comprises predicting whether the first service will be used by the user in the future time slot. The method comprises predicting, based on whether the first service has been predicted to be used in the future time slot, whether the second service will be used by the user in said future time slot.