Inferring User Affiliations via Social Graph Random Walks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional advertising targeting methods in social networking systems rely on incomplete and inaccurate demographic data, limiting the ability to effectively target users based on their affiliations and attributes, leading to inefficient ad spend and low click-through rates.
Innovation Solution
A social networking system determines seed clusters of users with specific attributes or affiliations by analyzing their connections, using explicit declarations, analysis of connected users, and random walk algorithms to infer affiliations, and then uses these clusters for targeted advertising, which is tested for accuracy through performance metrics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional demographic data and self-declared interests are used for advertising targeting, then the targeting process is simple and fast, but the accuracy and relevance of ad delivery is low
Solution Approach 1:
The patent replaces traditional mechanical demographic-based targeting with a computational inference system that analyzes social connection graphs. The system uses algorithms to infer user attributes, affiliations, and interests based on network relationships, substituting simple demographic filtering with complex but more accurate social network analysis to resolve the contradiction between targeting accuracy and system complexity
Solution Approach 2:
The patent introduces social connection data as an intermediary layer between users and advertisers. Instead of directly using self-declared user attributes, the system mediates through analysis of users' social networks, connections, and interactions to infer more accurate targeting characteristics, thereby improving precision while maintaining manageable complexity through structured analysis
2Loss of information
If advertisers purchase third-party analytical data to track user intent, then more information about user interests is obtained, but the cost of advertising increases and data accuracy remains uncertain
Solution Approach 1:
The patent enables the social networking system to self-generate valuable advertising targeting data by analyzing its own user interaction and connection data. Instead of purchasing external data, the system leverages its inherent social graph and user behavior information to create accurate user profiles and intent indicators, eliminating the need for expensive third-party data purchases while maintaining or improving information quality
Solution Approach 2:
The system uses feedback from user interactions, connections, and social network patterns to continuously refine and improve targeting accuracy. By monitoring how users interact with content, who they connect with, and their social network characteristics, the system generates increasingly accurate intent signals without requiring additional external data purchases, thereby reducing advertising spend while maintaining information quality
3Loss of information
If social networking systems passively record user information, then comprehensive user data is collected, but the system lacks tools to effectively use this data for advertising targeting
Solution Approach 1:
The patent applies preliminary action by pre-processing and structuring user social connection data into analyzable formats before advertising campaigns begin. The system pre-computes user profiles, connection graphs, and attribute inferences from passive user data, creating ready-to-use targeting segments that advertisers can immediately utilize, thereby transforming raw passive data into actionable targeting capabilities
Solution Approach 2:
The system replaces the lack of targeting tools with automated computational analysis capabilities. By implementing algorithms that automatically analyze social connection patterns, infer user attributes, and generate targeting segments, the patent substitutes the manual or non-existent targeting process with an automated system that effectively utilizes the passively collected user information
Data Source
AI summary
A seed cluster comprising a group of users who share a particular attribute and/or affiliation is determined by a social networking system. For each user of the seed cluster, other users and/or entities connected to the user in the social networking system are retrieved. For each retrieved other user or entity, the social networking system may determine whether the other user or entity exhibits the attribute or affiliation based on a random walk algorithm. A resulting targeting cluster of users and/or entities may be used for targeting advertisements targeting to members. A social networking system may also infer an affiliation for a user based on the user's interaction with a page, application, or entity where other users who interacted with the same page, application, or entity have the same affiliation.