User Clustering for Cloud Search Latency Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Multi-tenant cloud platforms face inefficiencies in handling search requests across different tenants due to varying user query histories, leading to increased querying latency and memory overhead.
Innovation Solution
Implementing a clustering system that groups clients with similar query histories using machine learning models, such as random forest models, to predict likely search entities and prioritize them in search results, thereby reducing latency and memory usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the cloud platform tracks and processes query histories for each user individually, then predictive accuracy for search results is improved, but memory overhead increases
Solution Approach 1:
The patent combines query histories from multiple users into cluster-level aggregated data structures. Instead of maintaining separate query history records for each user, the system groups users into clusters based on similar search patterns and maintains a single aggregated query history per cluster, thereby reducing memory overhead while preserving predictive accuracy through collective behavior patterns.
Solution Approach 2:
The patent segments the user base into distinct clusters based on query pattern similarities. By dividing users into groups with homogeneous search behaviors, the system can maintain optimized query histories for each segment rather than individual user histories, reducing overall memory requirements while maintaining prediction accuracy within each segment.
2Speed
If the system processes search requests without optimization, then device complexity is reduced, but querying latency increases
Solution Approach 1:
The patent performs preliminary clustering of users and pre-computation of query patterns before actual search requests are processed. By pre-grouping users into clusters and pre-analyzing query histories at the cluster level, the system reduces the computational burden during actual search operations, thereby reducing querying latency without requiring complex real-time processing mechanisms.
Solution Approach 2:
The patent introduces cluster-level aggregated query histories as an intermediary data structure between individual user queries and the search engine. This intermediary layer allows the system to serve multiple users from pre-computed cluster patterns, reducing the need for complex individual query processing while maintaining fast response times.
3Adaptability or versatility
If the platform serves multiple tenants with different search patterns, then adaptability is improved, but handling efficiency deteriorates due to varying query histories
Solution Approach 1:
The patent implements dynamic cluster formation and dissolution based on query pattern evolution. As users' search behaviors change over time, the system dynamically reassigns users to different clusters, allowing the platform to adapt to changing multi-tenant requirements while maintaining handling efficiency through continuously optimized cluster assignments.
Solution Approach 2:
The patent creates a universal clustering mechanism that serves multiple tenants with different search patterns using a single unified approach. The cluster-based architecture provides a multi-functional solution that can handle diverse query patterns across different tenants while maintaining consistent handling efficiency through standardized cluster management processes.
Data Source
AI summary
Methods, systems, and devices for user clustering on a cloud platform are described. A user associated with a client may implement a search function to query objects in a database system, and may select an entity (i.e., the clicked entity) from the search results. Each client may utilize the cloud platform in a specific manner, where users associated with the client may frequently search for certain types of objects. In some cases, one or more clients may share similar search histories or clicked entities. A clustering server may group clients with similar search histories or click distributions into common clusters. For future searches, the clustering server may utilize a machine learning model to predict the type of object being searched for based on the clustering. For example, user devices associated with a particular cluster may display similar groups and orders of object types in response to similar queries.