Ad Click Quality Scoring Using Conversion Probability Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional ad quality assessment systems are inflexible, oversimplify data, biased, and struggle with scalability and labor-intensive manual updates, leading to inaccurate and suboptimal ad performance in dynamic environments.
Innovation Solution
A machine learning-based interaction system that analyzes a broad range of user interaction metrics, continuously learns from new data, and adapts to changing behaviors, incorporating real-time data for nuanced and accurate ad click quality assessments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional rule-based systems are used for ad quality assessment, then the system structure is simple and easy to implement, but the system lacks flexibility and adaptability to changing user behaviors
Solution Approach 1:
The patent replaces traditional rule-based mechanical assessment systems with a machine learning-based intelligent system. The machine learning model automatically learns patterns from user interaction data without requiring manual rule configuration, enabling the system to adapt to changing user behaviors while maintaining operational simplicity through automated decision-making.
Solution Approach 2:
The machine learning system performs self-learning and self-adjustment by continuously processing user interaction data and updating its internal models. This self-service capability allows the system to automatically adapt to new user behaviors and patterns without requiring external intervention or manual rule updates, resolving the contradiction between adaptability and complexity.
2Productivity
If traditional rule-based assessment methods are used, then the implementation is straightforward, but the systems are labor-intensive and require manual updates
Solution Approach 1:
The patent replaces manual rule-based assessment processes with automated machine learning systems that continuously process and analyze user interaction data. This substitution eliminates labor-intensive manual updates and rule maintenance, significantly improving productivity through fully automated real-time ad quality assessment.
Solution Approach 2:
The machine learning system operates continuously to assess ad quality, processing user interaction data in real-time without interruption. This continuous automated operation eliminates the discontinuous manual update cycles of traditional systems, maintaining constant assessment efficiency and eliminating the need for periodic manual interventions.
3Measurement precision
If traditional ad quality assessment systems are used, then the system design is simple, but the systems produce inaccurate and biased assessments
Solution Approach 1:
The patent replaces simple but inaccurate rule-based assessment mechanisms with sophisticated machine learning models that analyze multiple user interaction features. This substitution improves measurement precision by capturing complex patterns and relationships in user behavior data, while the automated nature of machine learning prevents human biases from affecting assessment accuracy.
Solution Approach 2:
The machine learning system combines multiple diverse user interaction features and data sources to form a comprehensive assessment model. This composite approach integrates various signals such as click behavior, viewing time, and engagement patterns, creating a more accurate and nuanced understanding of ad quality that overcomes the limitations of single-metric traditional systems.
4Productivity
If traditional assessment systems are used, then the system is easy to maintain, but the systems struggle with scalability
Solution Approach 1:
The patent replaces manual maintenance processes with automated machine learning systems that scale efficiently with increasing data volumes and user interactions. The machine learning infrastructure is designed to handle large-scale data processing automatically, enabling the system to maintain high productivity and accuracy even as the platform grows, without requiring proportional increases in manual maintenance resources.
Data Source
AI summary
Described is a system for inferring ad quality by accessing user interaction data of users on an application and corresponding conversion data, the user interaction data including click behavior of the plurality of users; processing the user interaction data and the conversion data via a machine learning model to train the machine learning model, the machine learning model being trained to infer a conversion probability based on new user interaction data; displaying an impression on a user interface of the application to a first user; determining that the first user has selected the displayed impression; accessing first user interaction data of a first user indicative of first user click behavior; processing the first user interaction data via the machine learning model to generate a conversion probability; and determining whether the selection of the first user of the displayed impression is a low-quality click based on the conversion probability.