Pre-computed User Interest Tags for Ad Relevance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional targeted ad placement techniques are inefficient due to the need for real-time data processing, which can lead to increased response times and poor ad relevance, as they do not effectively account for individual user interests and often display the same ads to all users.
Innovation Solution
Determining recommendation data, such as ads, in advance by analyzing user historical action logs and assigning weight values to keywords based on recency, allowing for pre-computed and stored recommendations to be pushed to users, reducing the need for real-time processing and improving ad relevance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If real-time determination of ad placement is performed in response to user requests, then ad relevance to user interests can be improved, but processing time and server load increase significantly
Solution Approach 1:
The system performs preliminary analysis of user historical behavior data to generate user profiles and interest tags in advance, storing them in a database. When a user requests ad placement, the system retrieves pre-computed user profiles and matches them with ad candidates, avoiding real-time analysis of raw historical data and significantly reducing response time while maintaining ad relevance
Solution Approach 2:
The system extracts key interest tags and user attributes from historical behavior data in advance, separating the complex analysis process from the real-time ad delivery process. This extraction allows the system to work with condensed user profiles rather than raw historical data during ad placement determination
2Measurement precision
If frequent real-time processing of user data is performed to maintain ad relevance, then ad accuracy can be improved, but server processing efficiency deteriorates
Solution Approach 1:
User profiles and interest tags are computed and updated in advance based on historical behavior data, storing the results in a database. This preliminary computation eliminates the need for frequent real-time analysis of raw user data, reducing server processing load while maintaining accurate ad matching through pre-computed user profiles
Solution Approach 2:
The system creates simplified copies of user data in the form of interest tags and profile attributes that capture essential user characteristics. These copied representations are much smaller and faster to process than the original historical behavior data, enabling efficient real-time ad matching without sacrificing accuracy
3Device complexity
If the same predetermined ad is displayed to all users, then system complexity is reduced, but ad effectiveness decreases due to lack of personalization
Solution Approach 1:
The system assigns different interest tags and profile attributes to different users based on their individual historical behavior patterns. This allows the ad selection process to be customized for each user locally, improving ad effectiveness by matching ads to individual user interests while keeping the overall system architecture relatively simple through standardized profile structures
Data Source
AI summary
Determining recommendation data is disclosed, including: extracting a first set of keywords from a set of user action logs that occurred prior to a predetermined time point and determining a weight value for at least one of the first set of keywords; extracting a second set of keywords from a set of user action logs that occurred subsequent to the predetermined time point and determining a weight value for at least one of the second set of keywords; merging at least a portion of the first set of keywords and at least a portion of the second set of keywords to obtain a third set of keywords and determining a weight value for at least one of the third set of keywords; matching the third set of keywords to a database of data that can potentially be recommended to a user; and in the event that a piece of data is determined to match at least one keyword from the third set of keywords, determine that the piece of data is to be recommended to the user.