Machine Learning Ad Scorer for Search Relevance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Ad selection algorithms face challenges in accurately understanding user intent behind queries, particularly for long tail searches, while also requiring low computing cost, latency, and high throughput to serve high volume traffic, and are hindered by the need for empirical parameter tuning and suboptimal relevance scoring methods like Cosine Similarity.
Innovation Solution
A machine learning-based system is employed for ad selection, using a WAND algorithm framework with a machine learned IR scorer to determine relevant ads by constructing a linear model that maximizes relevance metrics, thereby eliminating the need for empirical parameter tuning and improving efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional ad selection algorithms use empirical parameter tuning and Cosine Similarity for relevance scoring, then the system can operate with simple implementation, but the measurement precision of ad relevance is insufficient
Solution Approach 1:
The patent replaces traditional mechanical similarity calculation methods (Cosine Similarity) with a machine learning-based relevance scoring system. The ML model learns optimal relevance patterns from historical data, substituting the rigid mathematical formula with an adaptive statistical model that achieves higher measurement precision without proportionally increasing system complexity.
Solution Approach 2:
The invention transforms the static parameter-based Cosine Similarity approach into a dynamic parameter learning system. Instead of fixing similarity weights manually, the system learns parameters from training data, allowing the scoring model to adapt to changing user behavior and ad patterns, thereby improving relevance measurement precision.
2Reliability
If ad selection algorithms perform comprehensive relevance evaluation to achieve high recall, then more relevant ads are identified, but the computing cost and latency increase
Solution Approach 1:
The system performs preliminary actions by pre-processing and indexing ad features before the actual selection query. Historical ad data is pre-trained and stored in an optimized format, allowing the ML model to quickly score new ads without performing comprehensive evaluations from scratch, thus maintaining high recall while reducing latency.
Solution Approach 2:
The ad selection process is segmented into multiple stages: initial filtering using lightweight features, followed by ML-based scoring for candidate ads, and final ranking. This segmentation allows comprehensive evaluation only for a subset of promising candidates, maintaining high recall while reducing overall computing cost and time loss.
3Measurement precision
If the system serves high volume traffic with accurate ad selection, then advertiser satisfaction improves, but the computing resources required increase
Solution Approach 1:
The ML model is trained offline on historical data to learn user intent patterns, making the online serving phase more efficient. The model serves itself by automatically scoring ads without requiring extensive real-time computing resources, thus improving measurement precision while reducing online energy consumption for high-volume traffic.
Solution Approach 2:
The system applies partial evaluation by focusing computing resources on the most promising candidate ads identified through initial filtering. Instead of exhaustively evaluating all ads for every query, the ML model scores only a subset of candidates, achieving high user intent understanding accuracy for served ads while minimizing overall computing resource consumption.
Data Source
AI summary
Methods, computer systems, and computer storage media are provided for evaluating information retrieval (IR) such as search query results (including advertisements) by a machine learning scorer. In an embodiment, a set of features is derived from a query and a machine learning algorithm is applied to construct a linear model of (query, ads) for scoring by maximizing a relevance metric. In an embodiment, the machine learned scorer is adapted for use with WAND algorithm based ad selection.