Ensemble Ranking Model for Influential User Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for identifying influential users in communication networks are inadequate, as single network measures are not predictive and supervised learning-based methods require labeled data, which is impractical for large-scale domains and domains with 'cold-start' issues.
Innovation Solution
A method that selects multiple ranking models based on usage data, calculates a weighing factor using an active learning process and stochastic gradient descent, and generates an aggregate score to identify influential users by comparing it to a predefined score.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single network measure (e.g., degree centrality, betweenness centrality) is used to identify influential users, then the identification process is simple and fast, but the accuracy and predictive power of identifying influential users is insufficient
Solution Approach 1:
The patent combines multiple network measures (degree centrality, betweenness centrality, closeness centrality, eigenvector centrality) and non-network measures (user attributes, usage patterns) into a unified ensemble model. This merging of multiple measurement approaches allows the system to capture different aspects of user influence simultaneously, thereby improving identification accuracy while managing complexity through systematic integration
2Measurement precision
If supervised learning-based methods are used to combine multiple centrality measures, then the predictive accuracy improves, but the requirement for large amounts of labeled data becomes a practical barrier
Solution Approach 1:
The patent performs preliminary actions by pre-computing multiple centrality measures and user attributes for all users in the network before the actual influential user identification task. This preprocessing step creates a rich feature set that can be leveraged by the ensemble model, reducing the need for extensive labeled training data while maintaining high predictive accuracy
Solution Approach 2:
The system uses unlabeled data from the network to compute centrality measures and user attributes, allowing the model to learn patterns from the inherent structure of the network data itself rather than relying entirely on externally labeled data. This self-service approach enables the system to leverage the abundance of available network data
3Measurement precision
If multiple ranking models and ensemble methods are employed to identify influential users, then the identification accuracy improves, but the computational complexity and processing time increase
Solution Approach 1:
The patent segments the identification process into distinct computational stages: (1) computing individual centrality measures for each user, (2) aggregating multiple measures into composite scores, and (3) applying the ensemble model to rank users. This segmentation allows each stage to be optimized independently and enables parallel computation of different centrality measures, reducing overall processing time while maintaining accuracy
4Adaptability or versatility
If traditional single-measure methods are used, then the system is easy to implement and interpret, but it cannot capture the multifaceted nature of user influence in communication networks
Solution Approach 1:
The patent creates a universal ensemble framework that can accommodate multiple types of measures (network-based centrality measures, attribute-based features, usage-based metrics) within a single unified model. This multi-functional system can adapt to different network types and influence definitions by simply changing the input measures, thereby capturing the multifaceted nature of user influence without requiring separate systems for each measure type
Data Source
AI summary
Embodiments of the invention disclose a method, identifier and a computer program product for identifying influential users among a group of users associated with a communication network. The disclosed method includes selecting two or more ranking models from a plurality of ranking models such that each ranking model provides scores to the users in the group based on a usage data of the users, calculating a weighing factor for each of the selected ranking models, generating an aggregate score for each user using the weighing factor and the score provided by each one of the selected model such that based on the aggregate score, the influential users among the group of users are identified.