Weighted Lookalike Audience Expansion for Online Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for defining online content audiences are inefficient, leading to content being delivered to users who are not valuable to content providers, resulting in wasteful use of time, screen space, and computing resources due to imprecise targeting.
Innovation Solution
An online system generates a weighted lookalike expansion of a custom-generated audience by allowing content providers to specify a subset of users with weights indicating their relevance, using a cluster model to identify similar users and expand the audience, thereby enabling more precise targeting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional audience definition methods are used, then content delivery simplicity is maintained, but audience targeting precision deteriorates
Solution Approach 1:
The audience definition process is segmented into multiple hierarchical levels: (1) seed user selection with individual weights, (2) cluster generation based on similarity metrics, (3) cluster weighting aggregation, and (4) threshold-based filtering. This segmentation allows precise targeting through weighted seed users while managing complexity through automated cluster-based expansion.
Solution Approach 2:
The system performs preliminary actions by pre-defining seed user groups with weights before actual content delivery. These weighted seed users serve as templates that automatically generate expanded audience clusters in advance, enabling precise targeting without real-time computation complexity during content distribution.
2Quantity of substance
If audience size is increased to reach more users, then content reach is improved, but resource waste increases due to including less valuable users
Solution Approach 1:
The system applies local quality by assigning different weights to different user clusters based on their similarity to weighted seed users. High-value users (those similar to weighted seeds) receive higher priority in content delivery, while lower-value users are included with reduced priority. This ensures expanded audience reach while maintaining resource efficiency through differentiated quality levels.
Solution Approach 2:
The system changes the parameter of user valuation from binary (included/excluded) to continuous (weighted scores). By computing similarity metrics and applying weight aggregation across clusters, the system creates a continuous spectrum of user value that allows precise control over resource allocation while expanding audience size to include marginally valuable users.
3Measurement precision
If weighted lookalike expansion is performed, then audience relevance is improved, but computational complexity increases
Solution Approach 1:
The system uses copying by creating user clusters that replicate the characteristics of weighted seed users. Instead of directly computing complex relationships with all users, the system copies the weighted profile of seed users to identify and group similar users, significantly reducing computational complexity while maintaining high audience relevance through similarity-based matching.
Solution Approach 2:
The system introduces cluster-based intermediaries as mediators between weighted seed users and the broader user population. These clusters serve as intermediate representations that aggregate user similarities, allowing the system to compute audience relevance through hierarchical clustering rather than direct pairwise comparisons, thereby reducing computational complexity while preserving precision.
Data Source
AI summary
An online system defines a new group of users of the online system for a content provider to use as an audience for its content. The online system generates the audience by first allowing the content provider to specify a subset of the users of the online system, along with a weight for each specified user. The online system then performs a lookalike expansion on the specified subset to obtain a larger group of users of the online system who are similar to the specified subset. This lookalike expansion is weighted using the weights provided for each of the members of the specified subset. Once the resulting expanded audience is generated, the online system performs a content selection process for content to present to users of the online system, where membership in the cluster group is used as an eligibility criteria in the content selection process.