Deep-Learning Account Tagging for Time-Series Event Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting user events, such as credit risk or response to offers, are limited by the difficulty in discovering meaningful parameter combinations, leading to inaccurate predictions and resource wastage.
Innovation Solution
A system using a neural network, like LSTM, processes historical user data in time series format to generate predictive values for events, allowing for more accurate tagging and decision-making by considering multiple parameters effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human-derived parameters and traditional models are used to predict user events, then the prediction process is simple and interpretable, but the prediction accuracy is insufficient due to inability to discover meaningful parameter combinations
Solution Approach 1:
The patent replaces traditional human-derived parameter models with a neural network-based deep learning system. The neural network automatically learns and extracts meaningful features from raw data without requiring manual parameter selection or combination, thereby substituting the mechanical process of human model building with an automated computational system that achieves superior prediction accuracy
Solution Approach 2:
The patent transforms the approach from using fixed human-derived parameters to dynamically learned parameters through neural network training. The system changes the parameter representation from static, pre-defined features to adaptive, data-driven features that are automatically optimized during training to maximize prediction accuracy
2Measurement precision
If more parameters and their combinations are considered to improve prediction accuracy, then the prediction becomes more comprehensive, but the difficulty of discovering meaningful combinations increases
Solution Approach 1:
The patent replaces the manual process of discovering parameter combinations with an automated neural network system. The neural network automatically explores the feature space and identifies meaningful combinations through its learning process, eliminating the difficulty of manual parameter combination discovery while comprehensively considering multiple parameters
Solution Approach 2:
The neural network performs self-service by automatically learning and extracting relevant features and their combinations from the data without human intervention. The system independently identifies meaningful parameter combinations through its training process, making the discovery process self-directed and eliminating the need for human expertise in feature engineering
3Productivity
If traditional prediction models are used, then resource consumption is lower, but inaccurate predictions lead to resource wastage through unnecessary communications and actions
Solution Approach 1:
The patent applies preliminary action by using the neural network to make accurate predictions before taking any resource-consuming actions such as sending communications or making business decisions. The high-accuracy predictions enable the system to pre-determine which actions are worthwhile, preventing resource wastage before it occurs
Solution Approach 2:
The system uses feedback from accurate neural network predictions to optimize resource allocation. By continuously learning from prediction outcomes and adjusting its model accordingly, the system improves its ability to identify which actions will be effective, thereby enhancing resource efficiency through iterative optimization
Data Source
AI summary
Methods and systems are presented for tagging an account associated with a user based on a predicted likelihood of an event associated with the user. A set of features is determined for data associated with the user. Values from the data are aggregated over time intervals for each feature to create time series data. The time series data is used as input to a neural network configured to accept input with the determined features. A predictive value indicating the likelihood of an event associated with the user is received from the neural network and used to determine whether to tag a user account. Determinations regarding the user are made based on the existence of absence of a tag on the user's account.