Hierarchical Clustering for Network User Behavior Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional clustering methods are inefficient in handling data points with multiple dimensions, often requiring excessive computational resources and time, and fail to provide accurate insights into user behavior on networks.
Innovation Solution
A hierarchical clustering method that uses a computationally efficient first pass clustering method like k-means to determine first-level cluster centroids, followed by additional passes to refine second-level and further cluster centroids based on average distance thresholds, iteratively until data points are closely grouped.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional clustering methods (e.g., k-means) are used for data points with multiple dimensions, then computational efficiency is improved, but clustering accuracy deteriorates
Solution Approach 1:
The patent segments the clustering process into multiple hierarchical levels (first-level cluster centroids, second-level cluster centroids, etc.). Each level performs clustering on a subset of data points with reduced dimensionality, achieving both computational efficiency and improved accuracy through progressive refinement of cluster assignments.
2Measurement precision
If more sophisticated clustering methods are used to improve clustering accuracy, then measurement precision is improved, but computing resources and time consumption increase
Solution Approach 1:
The patent divides the sophisticated clustering task into multiple simpler stages. First-level centroids are computed efficiently, then second-level centroids refine specific subsets. This segmentation achieves high accuracy without requiring a single computationally intensive clustering pass on all data.
Solution Approach 2:
The patent performs preliminary clustering to identify first-level centroids before conducting more refined clustering on specific subsets. This preliminary action reduces the scope of subsequent sophisticated clustering operations, saving time while maintaining accuracy.
3Measurement precision
If more sophisticated clustering methods are used to improve clustering accuracy, then measurement precision is improved, but computing resources increase
Solution Approach 1:
The patent segments the data processing into hierarchical levels where each level processes a subset of data points. This reduces the computational burden at each stage compared to applying sophisticated clustering methods to the entire dataset, while still achieving high clustering accuracy through cumulative refinement.
Data Source
AI summary
In one embodiment, a method includes receiving first network usage data for user devices indicating activity by the user devices on webpages, determining first-level cluster centroids based on the first network usage data, and assigning each user device to at least one first-level cluster centroid. The method also includes determining second-level cluster centroids based on the first network usage data for user devices assigned to a particular first-level cluster centroid, and assigning each user device previously assigned to the particular first-level cluster centroid to at least one second-level cluster centroid. The method further includes determining clusters based on the first-level cluster centroids and the second-level cluster centroids, assigning each user device to at least one cluster, and sending one or more first notifications to user devices associated with a particular cluster.