Intent Bias Model for Content Relevance Normalization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional approaches to displaying content in electronic environments, such as electronic marketplaces, do not adequately account for intent bias and position bias, leading to sub-optimal search results and increased user effort in finding relevant content.
Innovation Solution
The use of a generative model, specifically a Poisson-Beta model, to estimate true relevance and intent-based click-through rates, which factors in user intent and position bias to normalize content rankings and improve display positioning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional relevance information is used to determine content display order, then the system is simple to implement, but the search results are sub-optimal and require additional user effort
Solution Approach 1:
The patent segments the relevance determination process into multiple independent components: position bias model, intent bias model, and generative model. Each component addresses a specific aspect of bias correction, allowing the system to improve accuracy through modular, manageable segments rather than a monolithic complex system.
Solution Approach 2:
The patent applies preliminary action by pre-training bias correction models using historical data before actual content ranking. The position bias model and intent bias model are trained in advance to capture systematic patterns, so that during content display, the system can directly apply these pre-computed corrections without real-time complex calculations.
2Productivity
If conventional performance metrics are used for content ranking, then the system requires fewer computational resources, but user experience deteriorates due to sub-optimal results
Solution Approach 1:
The patent introduces intermediary models (position bias model and intent bias model) that mediate between raw performance metrics and final content ranking. These intermediaries correct systematic biases in the data before ranking, improving content discovery efficiency without requiring complete recalculation of all metrics from scratch.
Solution Approach 2:
The patent uses copying by training models on historical data copies and using these trained models to infer corrections for new data. The generative model copies the bias patterns learned from historical interactions and applies them to normalize new content rankings, reducing computational resources needed for real-time analysis.
3Ease of operation
If additional data is transmitted and displayed to help users locate content, then user effort increases, but the system requires more resources for transmission and display
Solution Approach 1:
The patent applies preliminary action by pre-processing and normalizing content rankings using trained bias correction models before data transmission. This ensures that the most relevant content is already positioned optimally in the results, reducing the need for users to scroll through or filter additional data, thereby decreasing the volume of data that needs to be transmitted and displayed.
Data Source
AI summary
The effect of intent bias on content performance can be determined in order to provide more relevant content in response to a query or other opportunity. Performance data can include the frequency with which an action, such as a purchase, occurs in response to an instance of the content being displayed. An intent bias model can be trained using the performance data for two or more intents, such as an action intent and an explore intent. Once the intent bias for an offer is determined, a normalized performance value can be obtained that does not include the effects of the bias. The normalized values can be used to select and place content based on actual performance.