Click-through Rate Model Using Cumulative Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Online systems face challenges in selecting relevant content for users while managing computational resources effectively, as they need to balance providing user-interacting content with the burden of processing large databases of pre-generated content items.
Innovation Solution
The online system employs a model serving system to determine scores for content items based on user interaction likelihood, relevance, and predicted click-through rates, using a cumulative scoring model that weighs features related to users and items, and ranks content for display based on these scores and bid prices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the online system retrieves and processes a large database of pre-generated content items to select relevant content for users, then the relevancy and user interaction likelihood of displayed content improves, but the computational resources and processing time required increases
Solution Approach 1:
The system pre-calculates and stores content scores for multiple content items before user requests arrive. When a user views an item, the system has already prepared relevance scores, interaction likelihood scores, and cumulative scores for candidate contents, eliminating the need for real-time computation and enabling instant content display
Solution Approach 2:
The content selection process is divided into independent scoring components: relevance scoring, interaction likelihood scoring, and cumulative score calculation. Each component processes content items separately and can be pre-computed, allowing the system to handle large databases efficiently by breaking down the complex selection task into manageable segments
2Adaptability or versatility
If the online system stores and processes a large database of pre-generated content items, then the ability to provide relevant content improves, but the computational resources required to store and search through the database increases
Solution Approach 1:
The system extracts only the most important scoring features from content items and user data, calculating cumulative scores based on weighted combinations of relevance and interaction likelihood. This extraction approach allows the system to work with a large database without processing all possible content attributes, reducing computational resource requirements while maintaining content selection quality
Solution Approach 2:
The system dynamically adjusts weighting parameters in the cumulative score calculation based on user behavior patterns and content performance metrics. By changing these parameters, the system can adapt to different user preferences and content types without restructuring the entire database or reprocessing all content items, optimizing resource usage while maintaining versatility
Data Source
AI summary
An online system receives an indication that a user is starting an order. The online system retrieves candidate contents for the user and provides prompts to a model serving system. The model serving system is configured to provide scores for the contents based on relevancy, a likelihood of user interaction, and a likelihood of the user purchasing an item associated with the content. The online system provides scores from the model serving system to a predicted click-through rate (pCTR) model. Based on the pCTR model scores, the online system ranks the candidate contents. The online system provides content for display to the user based on the ranked candidate contents.