Distributed Ad Ranking for Server Load Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional technologies face high server loads when displaying advertisements that consider user attributes and real-time viewing or purchase histories, especially during high web page request volumes, making it difficult to provide effective ad serving.
Innovation Solution
A distributed processing system where a processor calculates a first index for display-candidate content on a server and a second index on a user terminal, using location information to select and display advertisements, reducing data transfer and server load by only sending text data representing content location.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If processing to choose advertisements with consideration of user attributes and real-time viewing/purchase histories is performed on the server, then advertisement relevance to user is improved, but server load increases significantly
Solution Approach 1:
The advertisement selection process is segmented into two parts: the server performs preliminary filtering and sends only candidate advertisement lists to user terminals, while the user terminal performs the final selection based on real-time user attributes and viewing histories. This segmentation reduces server load while maintaining advertisement relevance.
Solution Approach 2:
The user terminal acts as an intermediary between the server and the advertisement selection process. It receives candidate lists from the server, applies local filtering based on user-specific data, and selects final advertisements without requiring the server to process all user attribute data.
2Productivity
If distributed processing is performed at the user terminal, then server load is reduced, but communication data amount increases
Solution Approach 1:
Only essential candidate advertisement lists are extracted and sent to user terminals, while detailed user attribute processing and final selection logic are extracted and executed locally at the user terminal. This minimizes communication data while enabling distributed processing.
3Speed
If all candidate advertisement data is sent to user terminals for local processing, then processing speed at terminal is improved, but communication load increases
Solution Approach 1:
Instead of sending all candidate advertisement data, only a partial list of promising candidates is sent to user terminals. The terminal performs sufficient processing on this reduced set to achieve good advertisement relevance without the need to process complete candidate sets, thus balancing communication load and processing speed.
Data Source
AI summary
An index calculation unit of a server calculates a Pre-Ad rank value that is a first index, on the basis of a GMS value and a unique users value. A list generation unit generates a list of display-candidate advertisements selected on the basis of the first indexes, the list including location information of the advertisements, and sends the list to a user terminal. A match level determination unit of the user terminal 3 calculates a correlation between a user and the product in relation to the user. An individual index calculation unit calculates an Ad rank value that is a second index, on the basis of the first index and the correlation. An advertisement choosing unit specifies a number of advertisements to be displayed and, on the basis of the second indexes, chooses the specified number of advertisements from among the advertisements of the one or more products.