Web Browsing Transformers for Predictive Targeting Under Data Limits
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing algorithms for predicting user behavior and selecting targeted content face challenges due to limited data availability and restrictions on data collection, making it difficult to create accurate user profiles for effective advertising.
Innovation Solution
Utilizing transformer models to analyze web browsing history and conversion events, incorporating a foundation model and prediction heads to predict missing websites or conversion probabilities, leveraging self-supervised tasks for training and applying attention mechanisms to establish relationships between websites.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional algorithms are used to predict user behavior, then the system can process data, but the prediction accuracy deteriorates due to limited data availability and shorter time periods
Solution Approach 1:
The patent introduces an intermediary representation layer that transforms raw browsing history into compressed user profiles and website embeddings. This intermediary representation captures essential patterns while reducing data volume requirements, allowing accurate predictions even with limited data availability.
Solution Approach 2:
The patent changes the parameter representation from raw browsing sequences to transformed embeddings and profiles. By converting browsing history into compressed representations with key patterns extracted, the system achieves better prediction accuracy with fewer data points.
2Measurement precision
If more data is collected to improve prediction accuracy, then the model performance improves, but the device complexity and data collection restrictions increase
Solution Approach 1:
The patent extracts only the essential patterns and features from browsing history rather than processing all raw data. By taking out key patterns and representing them in compressed form, the system achieves accurate predictions without requiring complex data collection and processing infrastructure.
Solution Approach 2:
The patent segments the data processing into distinct components: browsing history extraction, pattern identification, representation generation, and prediction. This segmentation allows each component to operate independently with reduced complexity, improving overall system efficiency.
3Reliability
If user profiles are created with limited data, then the system can operate within regulations, but the profile completeness and prediction reliability decrease
Solution Approach 1:
The system performs self-service by automatically extracting patterns and generating representations from the available browsing data without requiring external intervention or additional data collection. This self-service approach maximizes the use of available data to create reliable profiles within regulatory constraints.
Solution Approach 2:
The patent incorporates feedback mechanisms where the model continuously refines its predictions based on observed user behavior patterns. This feedback loop allows the system to improve prediction reliability even when working with limited and incomplete user profiles.
Data Source
AI summary
Some embodiments described herein relate to computer implemented methods and/or memory (e.g., a non-transitory, processor-readable medium) storing code configured to be executed by a processor to cause the processor to perform the method. The method can include training a first transformer with a plurality of website visitation records to predict a missing website in a sequence of websites. The first transformer can include a foundation model and a first prediction head. A second transformer model can be trained with a plurality of conversion event data to predict a probability of a conversion event occurring. The second transformer model can include the foundation model and a second prediction head.