Server Recommendation System Using Precomputed Ranked Lists
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional recommendation systems often require significant time to produce relevant content recommendations and may not provide content that is highly relevant to individual users.
Innovation Solution
A method and apparatus for generating a recommended subset of items for a user, where a server acquires user events and generates ranked predicted items lists using a ranking model algorithm, assigning item scores based on associated ranks, and selects items for a recommended subset based on these scores, allowing for real-time or offline generation and display of relevant content.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional recommendation systems use machine learning algorithms to generate content recommendations, then the relevance of recommended content to users is improved, but the time required to generate recommendations increases significantly
Solution Approach 1:
The system pre-generates ranked predicted items lists for multiple possible user queries before actual user requests arrive. These lists are stored and can be quickly retrieved and combined when a user query is received, eliminating the need to perform full machine learning ranking at request time while maintaining high relevance
Solution Approach 2:
The recommendation generation process is divided into two phases: an offline phase where ranked lists are pre-computed for various queries, and an online phase where these pre-computed lists are efficiently combined and merged. This segmentation allows computationally intensive operations to be performed in advance when time is not constrained
2Adaptability or versatility
If the system processes multiple user queries to generate comprehensive recommendations, then the coverage and relevance of recommendations are improved, but the computational complexity and processing time increase
Solution Approach 1:
The system generates ranked predicted items lists for more queries than any single user is likely to submit (excessive action), and stores these pre-computed results. When a user query arrives, the system only needs to combine and filter the relevant pre-computed lists rather than generating everything from scratch, reducing online computational complexity
Solution Approach 2:
By pre-computing and storing ranked items lists for multiple queries in advance, the system transforms the computationally intensive query processing into a simpler list combination and filtering operation at request time, managing complexity through advance preparation
Data Source
AI summary
A method of generating a recommended subset of items for a user of an electronic device, the method being executed at a server, the method comprises: acquiring user events associated with a plurality of users, the user events comprising indications of user queries; for each of the user queries, generating a ranked predicted items list that comprises at least some items from a set of potentially recommendable items, such that each particular item within the ranked predicted items list has an associated rank; for each item within a plurality of ranked predicted items lists, generating, by the server, an item score based on a totality of ranks associated therewith; generating the recommended subset of items from the set of potentially recommendable items by selecting at least one item within the plurality of ranked predicted items lists as the recommended subset of items based on the item scores.