Hierarchical Machine Learning Models for Search Result Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional search engines in multi-tenant online systems provide inadequate user experiences as they fail to tailor search result rankings to individual user preferences and organizational priorities, leading to either mediocre or poor user experiences due to the use of a single search engine model for diverse user needs.
Innovation Solution
Implementing a hierarchy of machine-learned models that select the appropriate ranking model based on dimensions such as organization and entity type associated with the search query, allowing for personalized ranking criteria and improving search result relevance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single search engine model is used for all users, then the system complexity is low and ease of operation is maintained, but the adaptability to different user preferences and organizational priorities deteriorates
Solution Approach 1:
The patent segments the search engine into multiple specialized models, each optimized for specific user preferences or organizational priorities. Instead of one general model, the system divides functionality into distinct ranking models that can be selectively applied based on user characteristics, thereby improving adaptability while managing complexity through modular design.
Solution Approach 2:
The system dynamically selects which search model to apply based on real-time user context and preferences. The model selection is not static but adapts to each user's needs, allowing the system to switch between different ranking strategies as required, thus achieving high adaptability without requiring all models to be active simultaneously.
2Adaptability or versatility
If multiple specialized ranking models are implemented to meet diverse user needs, then adaptability to different user preferences improves, but the device complexity and difficulty of operation increase
Solution Approach 1:
The system automatically performs model selection without requiring user intervention. Users simply submit their search queries, and the system autonomously determines which specialized model to apply based on user profiles, organizational context, and query characteristics. This self-service approach maintains ease of use while leveraging multiple specialized models.
Solution Approach 2:
The patent introduces an intermediary layer (model selection mechanism) that sits between the user and the multiple specialized models. This intermediary automatically matches user requests with appropriate models, shielding users from the complexity of multiple models while still benefiting from their specialized capabilities.
3Measurement precision
If conventional ranking criteria are used, then the system is simple to operate, but the precision of search results deteriorates because different users have different priorities for recall and precision
Solution Approach 1:
Different parts of the system (different ranking models) are optimized for different local requirements. Some models prioritize precision for users who need accurate results, while others prioritize recall for users who need comprehensive coverage. Each model has specialized quality characteristics suited to its intended use case, rather than attempting to optimize all models for all criteria.
Data Source
AI summary
A multi-tenant system stores a hierarchy of machine-learned models, wherein each machine-learned model is configured to receive as input a set of search results and generate as output scores for ranking the set of search results. Each machine-learned model is associated with a set of dimensions. The system evaluates search query performance. Performance below a threshold causes a new model to be generated and added to the hierarchy of models. Upon execution of a new search query associated with the same set of dimensions as the newly created model, the new model is used to rank that search query's search results.