Document Score Adjustment via Attractiveness Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing document scoring systems in search engines rely on handcrafted rules for score adjustments, which are computationally inefficient and difficult to apply across a range of documents, often resulting in contradictory outcomes due to the need for significant domain knowledge.
Innovation Solution
A computer-implemented method for determining score adjustments using user interaction data to train an adjustment model, allowing for automated and computationally efficient ranking of documents by learning from user preferences through attractiveness models and pairwise comparisons.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If handcrafted rules are used for document score adjustments, then domain expertise can be applied to improve ranking accuracy, but the system becomes computationally inefficient and difficult to apply across a range of documents
Solution Approach 1:
The patent replaces the mechanical system of handcrafted rules with a machine learning model (neural network) that automatically learns document attractiveness from user interaction data. This substitution eliminates the need for manual rule creation while improving computational efficiency through automated pattern recognition.
Solution Approach 2:
The system enables self-service by allowing the machine learning model to automatically learn and improve document ranking based on user interactions without requiring continuous manual intervention from search relevance engineers. The model trains itself on pairwise comparisons and user feedback data.
2Adaptability or versatility
If handcrafted rules are used for document score adjustments, then specific domain knowledge can be incorporated, but the system requires significant domain knowledge and can result in contradictory rules
Solution Approach 1:
The patent replaces complex handcrafted rules with a machine learning model that automatically learns from data. This eliminates the need for domain experts to manually create and maintain complex rule sets, reducing system complexity while maintaining adaptability through data-driven learning.
Solution Approach 2:
The system incorporates feedback loops where user interactions with search results are continuously collected and used to train and refine the machine learning model. This feedback mechanism allows the system to adapt to changing user preferences without manual rule updates.
3Measurement precision
If multiple comparisons are performed to determine document attractiveness, then more accurate ranking can be achieved, but the computational cost increases significantly
Solution Approach 1:
The patent performs preliminary action by pre-training the machine learning model on extensive pairwise comparison data before actual search operations. This pre-computed knowledge is then applied efficiently during search without requiring multiple real-time comparisons, reducing computational resource consumption while maintaining accuracy.
Data Source
AI summary
Disclosed is a computer-implemented method for determining a score adjustment for a search document, comprising determining a first attractiveness model of a first document from one or more documents based on one or more user interactions associated with the first document; determining a second attractiveness model of a second document from one or more documents based on one or more user interactions associated with the second document; determining one or more pairwise comparisons of documents based on the first and second attractiveness models of the first and second documents; training an adjustment model based on the pairwise comparisons of documents; and inputting the search document into the adjustment model to determine the score adjustment.