Audience Targeting Search with Relevance-Variety Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audience targeting systems prioritize relevance over variety, resulting in similar search results, which limits the diversity of target audiences for marketing campaigns.
Innovation Solution
A system and method that employs semantic distances and multiscale clustering to control both relevance and variety in search results, using a seed-grow-balance approach with language models and cosine similarity calculations to optimize target audience selection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing systems use traditional ranking metrics (TF-IDF, PageRank) to prioritize relevant results, then relevance is improved, but variety deteriorates
Solution Approach 1:
The system changes the ranking parameter from traditional metrics (TF-IDF, PageRank) to a new variety metric based on semantic distance and clustering. This parameter change allows the system to evaluate results based on their diversity in the solution space rather than just their relevance score, thereby improving variety while maintaining relevance through the seed-grow-balance approach
Solution Approach 2:
The patent introduces an intermediary mechanism (multiscale clustering based on semantic distance) that acts as a mediator between the search query and the ranking process. This intermediary structure organizes results into clusters and selects representatives from different clusters, ensuring both relevance (through semantic matching) and variety (through cluster diversity) in the final results
2Adaptability or versatility
If semantic distance calculation and multiscale clustering are implemented to improve variety, then variety is improved, but computational load increases
Solution Approach 1:
The system segments the search results into multiple clusters based on semantic distance. By dividing the results into distinct clusters and selecting representatives from each cluster, the system achieves variety without needing to compute and compare all possible result combinations, thereby reducing computational load while maintaining diversity
Solution Approach 2:
The patent implements a dynamic seed-grow-balance approach where the system adaptively adjusts the number of clusters and selection criteria based on the specific query and result set. This dynamic adjustment allows the system to optimize computational resources by processing only the necessary number of clusters and results, reducing overall computational load while maintaining variety
Data Source
AI summary
Disclosed herein is a system and method for controlling relevance and variety in stored data in data management platforms for audience targeting. The method comprises receiving an input query using a user interface from a user; storing and indexing the target audience in a database; providing an expanded set of search queries from the input query based on the input query, by using a language model; receiving the expanded set of search queries to provide a result of matched target audiences based on the expanded set of search queries matched with the target audiences stored in the database; receiving matched target audiences from a relevance module to provide a filtered result of the matched target audiences by discarding non-relevant target audience based on preset calculation; and performing multiscale clustering to provide a graphical representation or a ranked list of the filtered matched target audiences based on a weightage score.