User Behavior Model for Real-Time Bidding Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In real-time bidding for digital advertising, the lack of user identifiers due to privacy concerns hinders the ability to effectively predict user objects, impacting advertising efficiency.
Innovation Solution
An advertisement server uses machine learning algorithms to build a user behavior model from telecommunication data, predicting a candidate user identifier based on advertisement identifiers even when user or advertisement-related identifiers are absent in bid requests.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If user identifiers are removed from bid requests due to privacy concerns, then user privacy is protected, but the ability to predict user objects and improve advertising effectiveness deteriorates
Solution Approach 1:
The patent introduces an intermediary mechanism - a user behavior model trained on telecommunication data - that mediates between privacy protection and advertising effectiveness. Instead of directly using user identifiers, the system uses the advertisement identifier as an intermediary to query the pre-trained model, which then predicts user object information without exposing actual user identifiers in the bid request
Solution Approach 2:
The system performs preliminary action by pre-training the user behavior model offline using telecommunication data before the actual real-time bidding process. This preprocessing step creates a predictive framework that can be quickly queried during RTB without requiring real-time access to user identifiers, thus protecting privacy while maintaining advertising effectiveness
2Reliability
If traditional identifier-based user prediction is used, then advertising effectiveness is improved, but user privacy is compromised
Solution Approach 1:
The patent replaces direct identifier usage with an intermediary prediction system. The advertisement identifier serves as a key to query the user behavior model, which acts as an intermediary layer that translates ad identifiers into predicted user object information without requiring actual user identifiers to be present or accessible
3Measurement precision
If machine learning models are trained on telecommunication data, then user prediction accuracy is improved, but data processing complexity increases
Solution Approach 1:
The complex task of training machine learning models on telecommunication data is performed as a preliminary offline action. The model is trained in advance using historical data, and the trained model is then deployed for quick inference during real-time bidding. This separates the computationally intensive training phase from the time-sensitive RTB phase, reducing real-time processing complexity
Solution Approach 2:
The system segments the overall prediction task into two distinct phases: offline model training using telecommunication data, and online prediction during RTB using the pre-trained model. This segmentation allows complex data processing to be performed when computational resources are abundant, while real-time operations use the pre-processed model
Data Source
AI summary
The disclosure provides a method for predicting a user object in a real time bidding and an advertisement server. The method includes: obtaining telecommunication data, and building a user behavior model based on the telecommunication data by using at least one of machine learning algorithms; receiving a bid request comprising an advertisement identifier which does not have at least one of an identifier field of a certain user object and an advertisement-related identifier from an advertising trading platform; and predicting a first candidate identifier of a first user object corresponding to the bid request according to the advertisement identifier through the user behavior model.