Group Discovery Algorithm for Social Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing friend networks are ineffective in discovering groups of people with common attributes beyond friendship, as they primarily rely on connectivity based on friendship, making it difficult to identify groups sharing other characteristics.
Innovation Solution
An algorithmic approach that constructs an adjacency list from a friend network, uses the Facebook API to gather friend information, and employs cluster detection techniques to identify groups with common attributes by analyzing 2-neighbors and calculating alpha and beta coefficients to merge similar clusters, facilitating the presentation of relevant groups to users.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If friend network uses only friendship connectivity to define groups, then the graph structure remains simple and easy to maintain, but it becomes ineffective in discovering groups with common attributes beyond friendship
Solution Approach 1:
The patent transitions from a single-dimension friendship graph to a multi-dimensional analysis by introducing auxiliary information (attributes, interests, demographics) as additional dimensions. This allows the system to discover groups based on common attributes beyond just friendship connections, resolving the contradiction between maintaining simplicity and enabling versatile group discovery.
Solution Approach 2:
The patent introduces an intermediary processing layer that connects the friendship graph with auxiliary information. This intermediary layer (the group discovery algorithm) mediates between the simple friendship structure and the complex attribute-based grouping, enabling versatile group discovery without fundamentally complicating the underlying graph structure.
2Measurement precision
If the system analyzes comprehensive friend information to discover groups, then group discovery accuracy improves, but the computational time and resources increase significantly
Solution Approach 1:
The patent segments the large-scale friend network into smaller, manageable components for analysis. By dividing the comprehensive friend information into discrete attributes and interests, the system can process and analyze groups more efficiently while maintaining high accuracy in group discovery, thus reducing computational time without sacrificing precision.
Solution Approach 2:
The patent changes parameters by focusing on specific attributes and interests rather than analyzing all possible friend information comprehensively. This parameter change allows the system to achieve high group discovery accuracy by targeting relevant dimensions, thereby reducing the overall computational time and resources required.
3Loss of information
If the friend network stores detailed auxiliary information about each person, then the ability to identify groups with common characteristics improves, but the data storage requirements and processing complexity increase
Solution Approach 1:
The patent implements a universal auxiliary information structure that serves multiple functions: storing attributes, interests, and demographic data. This multi-functional data structure enables comprehensive group discovery while avoiding the need for separate complex structures for each type of information, thus reducing overall data structure complexity while maintaining full information availability.
Data Source
AI summary
A method for discovering a group defined by a common characteristic is disclosed. The method includes building representation of a portion of a social network based on a starting person with the given characteristic, the person also providing the person's gender and school affiliation. The social network representation is then searched to discover clusters therein meeting certain size and connectivity requirements with respect to the network. After the clusters in the network are discovered, clusters having a high degree of similarity are merged together. The resulting clusters, both merged and non-merged, are then scored to determine the cluster that best fits the original group. The winning cluster is then returned to the starting person who confirms the correctness of the cluster. The set of the persons in a confirmed cluster are then displayed to the starting person.