Conversion Prediction Model for Dynamic Content Bidding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing content serving platforms struggle to accurately and precisely select content items that will be of interest to users, leading to suboptimal user engagement and conversion rates.
Innovation Solution
A model is generated and trained to predict conversion-given-click probabilities of users and products, using positive and negative events to incrementally improve its accuracy. This model is then used to generate bids for products, determining their eligibility to compete in auctions hosted by the content serving platform.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a content serving platform uses traditional content selection methods, then the system complexity remains low, but the accuracy of content selection and user engagement is insufficient
Solution Approach 1:
The system performs preliminary training of the conversion prediction model using historical event data before actual content serving. This pre-computation of conversion probabilities enables accurate content selection during runtime without adding complex real-time processing, resolving the contradiction between selection accuracy and system complexity
Solution Approach 2:
A conversion prediction model is introduced as an intermediary component between the content serving platform and user engagement metrics. This model translates historical event data into conversion probabilities, enabling accurate content selection without directly increasing the complexity of the core serving infrastructure
2Measurement precision
If the platform trains the model on extensive positive and negative events, then the prediction accuracy improves, but the training time and computational resources increase
Solution Approach 1:
The model training is performed in advance using accumulated historical event data, converting unstructured event logs into a trained prediction model before deployment. This pre-computation approach achieves high prediction accuracy without incurring training time delays during actual content serving operations
Solution Approach 2:
The system uses a substantial amount of historical event data (excessive action) for training to ensure high prediction accuracy, but performs this training once in advance rather than continuously. This partial training approach maintains accuracy while avoiding ongoing time losses
3Productivity
If the platform generates bids based on conversion-given-click probability, then the conversion rate improves, but the computational overhead for bid generation increases
Solution Approach 1:
The conversion prediction model computes conversion probabilities in advance during the training phase, storing these probabilities for rapid retrieval during bid generation. This pre-computation reduces the computational overhead at bid generation time while maintaining high conversion rates
Solution Approach 2:
The prediction model serves itself by generating conversion probabilities that are directly used in bid calculations. The model's output automatically feeds into the bidding mechanism, eliminating the need for separate complex computational processes and reducing overall computational overhead
Data Source
AI summary
One or more systems and/or methods for providing conversion prospecting for dynamic content recommendation are provided. A model is trained, using positive events and negative events, to predict conversion-given-click probabilities of users and products. The model is utilized to generate a prediction of a conversion-given-click probability that a user will perform an action in relation to a product. The model is used to generate a bid for the user and the product based upon the conversion-given-click probability, a content provider bid for the product, and a target cost per action. The bid is used to determine whether the product is to compete in an auction hosted by a content serving platform that selects and transmits content items of products to devices for display to users.