Ad Conversion Prediction Using Multi-Task Learning Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Selecting advertisements for transmission to diverse devices at various locations is challenging due to the difficulty in predicting conversion rates effectively.
Innovation Solution
An advertising system that utilizes a machine learning model trained on multi-task learning to predict conversion rates based on pixel events and click-through and view-through conversions, applying inverse propensity weighting to select advertisements for display.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional advertisement selection methods are used, then the system is simple to operate, but the conversion rate prediction accuracy is low
Solution Approach 1:
The patent introduces a machine learning model as an intermediary component between the advertisement selection system and the conversion rate prediction task. This model processes pixel events and bid features to generate accurate conversion rate predictions, resolving the contradiction by adding intelligence (the model) rather than complex manual procedures
Solution Approach 2:
The patent replaces traditional mechanical/manual advertisement selection methods with an automated machine learning-based system. The model automatically processes features, performs inverse propensity weighting, and selects advertisements, substituting manual operations with intelligent automation that improves accuracy without proportionally increasing operational complexity
2Measurement precision
If multiple conversion metrics are considered, then the advertisement selection accuracy improves, but the computational complexity increases
Solution Approach 1:
The patent performs preliminary actions by pre-processing pixel events into standardized features and pre-training the machine learning model on historical data. This preparation work is done beforehand, allowing the system to efficiently handle multiple conversion metrics during actual advertisement selection without excessive computational overhead in real-time operations
Solution Approach 2:
The patent applies inverse propensity weighting to transform the probability distribution of pixel events, changing the parameters of the input data to account for selection bias. This parameter transformation allows the model to accurately process multiple conversion metrics while maintaining computational efficiency through mathematical optimization rather than brute-force computation
3Measurement precision
If inverse propensity weighting is applied, then the conversion rate prediction becomes more accurate, but the processing time increases
Solution Approach 1:
The patent calculates propensity scores and applies inverse propensity weighting in advance during the model training phase rather than during real-time advertisement selection. This preliminary computation of weighting factors allows the system to achieve accurate conversion rate predictions without incurring excessive processing time during actual bid evaluation
Solution Approach 2:
The patent applies inverse propensity weighting selectively to the most critical features and pixel events that have the greatest impact on conversion rate prediction accuracy. By focusing the computational effort on the most influential parameters rather than uniformly processing all data, the system achieves high accuracy while minimizing processing time
Data Source
AI summary
Aspects of the present disclosure involve a system comprising a storage medium storing a program and method for predicting a conversion rate. The program and method provide for receiving, from an advertisement service, a bid to display a first advertisement at a computing device; determining, in response to receiving the bid, a set of features that relate to the first advertisement; providing the set of features to a machine learning model configured to output a predicted conversion rate for the first advertisement, the machine learning model having been trained based on multi-task learning using plural sets of features corresponding to plural second advertisements, the plural sets of features being associated with both click-through conversions and view-through conversions; and determining, based on the output of the machine learning model with respect to the set of features, the predicted conversion rate for the first advertisement.