Ad Targeting via Segmented Candidate Generation and ML Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing information management systems face inefficiencies in selecting relevant content items, particularly in keyword-less ad targeting, where ads are not matched with query keywords, leading to suboptimal ad selection and user experience.
Innovation Solution
A system and method that determine base scores for content items based on attributes, adjust scores based on query-content item relationships using an odds model, and select the highest scoring items, enabling efficient keywordless ad targeting and improved user experiences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If keyword-less ad targeting is used to improve ad selection relevance, then ad relevance to user intent is improved, but computational complexity and processing time increase significantly
Solution Approach 1:
The patent segments the ad selection process into two distinct phases: (1) candidate generation phase that uses keyword matching to quickly filter ads, and (2) ranking phase that uses machine learning models to rank candidates. This segmentation allows the system to maintain high relevance through ML-based keyword-less targeting while managing computational complexity by limiting ML processing to a smaller candidate set rather than the entire ad inventory.
Solution Approach 2:
The patent performs preliminary keyword-based filtering to generate a candidate ad set before applying computationally intensive machine learning models. This preliminary action reduces the scope of subsequent ML processing, thereby managing computational complexity while still enabling keyword-less relevant ad selection through the two-stage approach.
2Measurement precision
If machine learning models are used to rank ads to improve selection accuracy, then ad selection accuracy is improved, but processing speed decreases
Solution Approach 1:
The patent divides the ad selection process into candidate generation (fast keyword matching) and ranking (accurate ML-based) phases. This segmentation allows the system to achieve high selection accuracy through ML models while maintaining processing speed by applying these models only to a filtered candidate set rather than the complete ad inventory.
Solution Approach 2:
The patent applies machine learning models selectively to only the top candidate ads generated by keyword matching, rather than processing all ads in the inventory. This partial application of ML processing achieves high selection accuracy for the most relevant ads while significantly reducing overall processing time and computational resources required.
3Measurement precision
If the system processes all content items to ensure optimal selection, then selection quality is improved, but processing time and computational resources increase
Solution Approach 1:
The patent extracts and processes only the most promising candidate ads for ML-based ranking, rather than processing all content items. By taking out the top candidates from the keyword-matched set and applying sophisticated ranking only to these, the system maintains high selection quality while dramatically reducing processing time and computational resource consumption.
Solution Approach 2:
The patent applies full ML-based ranking processing only to a partial set of candidate ads (those that passed keyword filtering), rather than processing the entire ad inventory. This partial processing approach ensures optimal selection quality for relevant ads while avoiding the excessive processing time that would result from analyzing all content items.
Data Source
AI summary
System and method for quickly selecting content items (e.g., documents) having highest ranking scores in response to a query for content items. Base scores for a plurality of content items are determined based on attributes of the content items. One or more attributes are extracted from the query, the base scores of at least some of the content items are adjusted based on relationships between the query attributes and the content item attributes. A subset of the plurality of content items having the highest adjusted scores are selected, and the selected subset of content items are output.