Multiscale Record Vectors for Cross-Session Behavior Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Enterprise-scale systems face challenges in predicting user behavior across different timescales due to the increasing number of user interface events, requiring a more robust and scalable approach to dimensionally reduce data for effective prediction and fraud detection.
Innovation Solution
A multi-scale approach using machine learning models, such as autoencoders and neural networks, to segment user interactions by digital sessions, generate record-characterizing vectors, and combine them across different timescales for predictive analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional dimensionality reduction methods are used to reduce user interface events, then data processing efficiency is improved, but prediction accuracy across different timescales deteriorates
Solution Approach 1:
The patent segments user interaction data into multiple timescale components (short-term, medium-term, long-term patterns) and processes each segment separately through dedicated neural network layers. This allows the system to maintain high processing efficiency while preserving prediction accuracy across different temporal granularities, resolving the contradiction between speed and precision.
Solution Approach 2:
The patent transforms temporal data into a multi-dimensional representation by adding the timescale dimension. Instead of reducing dimensions, it expands the data into multiple dimensional layers representing different time horizons, enabling simultaneous efficient processing and accurate multi-timescale prediction.
2Loss of information
If all user interface events are collected and processed, then prediction completeness is improved, but data volume and processing complexity increase
Solution Approach 1:
The patent extracts only the essential predictive features from the vast amount of user interface events using automated feature engineering. It identifies and extracts key interaction patterns, ignoring redundant data, thereby maintaining prediction completeness while significantly reducing processing complexity and data volume.
Solution Approach 2:
The patent replaces manual feature engineering and complex processing rules with automated neural network-based feature extraction. This substitution of mechanical processing with intelligent automation reduces processing complexity while preserving or enhancing prediction completeness.
3Quantity of substance
If data is reduced to lower dimensions, then storage and processing requirements are reduced, but the ability to detect complex fraud patterns deteriorates
Solution Approach 1:
The patent creates a composite representation by combining multiple feature types (demographic features, behavioral features, temporal features) into an integrated low-dimensional vector. This composite structure preserves fraud detection capability by maintaining the synergistic relationships between different feature types while reducing overall data volume.
Solution Approach 2:
The patent transforms high-dimensional raw data into optimized low-dimensional parameters through neural network encoding. The parameter transformation preserves critical fraud-related information by learning optimal parameter representations that maintain detection sensitivity while reducing data volume for efficient storage and processing.
Data Source
AI summary
A method includes segmenting updates associated with a record into a set of update subsets and generating first and second vectors based on first and second update subsets using a first neural network. The first update subset is associated with a first session and a timestamp, and the second update subset is associated with a second session. The method includes determining a first output using a second neural network based on the first and second vectors and a time difference between the first and second sessions. The method includes selecting a segment of a periodic time interval based on the timestamp, determining a second output using a third neural network based on a ratio based on the segment and the periodic time interval, and generating a characterizing vector using a fourth neural network based on the first and second outputs.