Weighted ML Model for Website-Specific Search Output
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing online store websites face suboptimal search output arrangements when using a single machine learning algorithm across multiple websites with different geolocations, as the algorithm fails to account for unique features and options of each website, leading to reduced user interaction probabilities.
Innovation Solution
A machine learning model is trained using data from multiple websites, with weights assigned based on similarities and differences between websites to prioritize item characteristics and user interactions specific to each website, ensuring accurate search output arrangements that maximize user interaction probabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single machine learning algorithm is used across multiple websites, then the device complexity is reduced, but the adaptability to different website characteristics deteriorates
Solution Approach 1:
The patent segments the training data from different websites by assigning weights to transactions based on website characteristics. Each website's transactions are weighted according to how well they match the target website's features, allowing the single algorithm to process multiple data sources while adapting to website-specific nuances through differential weighting rather than through multiple separate algorithms
Solution Approach 2:
The patent changes the parameter of data weighting dynamically based on website characteristics. By adjusting the weights assigned to training transactions according to website similarity metrics, the system adapts the algorithm's behavior to different website contexts without requiring separate algorithms for each website
2Adaptability or versatility
If website-specific machine learning algorithms are trained for each website, then the adaptability to website characteristics is improved, but the device complexity and data requirements increase
Solution Approach 1:
The patent creates a universal machine learning algorithm that can function across multiple websites simultaneously. By designing a single algorithm that processes weighted training data from various websites, the system achieves multi-functionality where one algorithm serves multiple websites with different characteristics, eliminating the need for separate algorithms for each website
3Productivity
If historical data is used to prioritize items, then the productivity of item presentation is improved, but the measurement precision of user interaction probability deteriorates due to irrelevant characteristics
Solution Approach 1:
The patent applies local quality by weighting training transactions according to website-specific characteristics. Transactions from websites with similar characteristics receive higher weights, while those with dissimilar characteristics receive lower weights. This ensures that the historical data used for prioritizing items is locally relevant to each website's context, improving the precision of user interaction probability measurements
Data Source
AI summary
Data from a first website and second website is used to determine a model for presenting a search output in response to a query received by the first website. The data from each website includes parameters of search queries, characteristics of items that were output in response, and an indication of items that were purchased. Characteristics of the first website are used to determine weights that are applied to data from the second website. A relationship between search queries received by each website is used to determine a first weight. A relationship between features or options offered by the second website and those offered by the first website is used to determine a second weight. For example, if the second website offers a service that the first website does not, the availability of this service for particular items may be disregarded when determining the model for the first website.