Customized Hidden Markov Models for Predictive Account Interaction Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems for predicting user account interactions with computing applications are inflexible, inaccurate, and inefficient, relying on rigid assumptions, requiring large datasets, and expending excessive resources, leading to unreliable predictions and inefficient resource allocation.
Innovation Solution
The use of customized hidden Markov models generated using neural networks to predict user account interactions, allowing for individualized predictions based on user account data, enabling flexible predictions across multiple time periods and applications with reduced data requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional systems utilize large datasets of historical data to generate predictions, then prediction coverage and completeness improve, but computing time, memory, storage space, and processing bandwidth requirements increase excessively
Solution Approach 1:
The patent extracts only the essential features from historical data using feature extraction techniques, transforming raw data into meaningful representations that capture user behavior patterns without requiring the entire historical dataset. This allows the system to generate accurate predictions while significantly reducing the amount of data that needs to be processed and stored.
Solution Approach 2:
The patent segments the prediction problem into multiple components by creating separate models for different aspects of user behavior (e.g., interaction frequency, interaction type, time-based patterns). Each model processes specific feature subsets independently, reducing the computational burden on any single model while maintaining overall prediction accuracy.
2Measurement precision
If conventional systems require years of historical account data to generate predictions, then prediction accuracy improves, but data preprocessing, cleaning, and maintenance requirements increase extensively
Solution Approach 1:
The patent performs preliminary feature extraction and transformation during the data collection phase, pre-processing data into meaningful representations before it is needed for prediction. This preliminary action reduces the complexity of subsequent data cleaning and maintenance operations while maintaining prediction accuracy.
Solution Approach 2:
The patent introduces intermediate representation layers between raw historical data and final predictions. These intermediate features serve as mediators that simplify the relationship between raw data and prediction targets, reducing the complexity of data processing while maintaining accurate predictions.
3Device complexity
If conventional systems utilize rigid assumptions and sub-models to predict account interactions, then model structure simplicity improves, but prediction robustness and flexibility deteriorate
Solution Approach 1:
The patent employs dynamic modeling techniques that allow the prediction model to adapt its structure and parameters based on the specific characteristics of each user account and the context of interactions. This dynamic approach enables the model to handle variability in account interactions without requiring rigid assumptions, improving both flexibility and robustness.
Solution Approach 2:
The patent applies different modeling approaches and feature sets to different aspects of user behavior and different user accounts. Each user account receives customized predictions based on its specific patterns, and different interaction types are modeled with appropriate local detail, maintaining overall model simplicity while enabling local adaptability.
4Device complexity
If conventional systems generate single prediction values for single applications and time periods, then prediction model simplicity improves, but prediction completeness and robustness across multiple time scales deteriorate
Solution Approach 1:
The patent extends predictions from single time-point forecasts to multi-time-scale predictions by adding temporal dimensionality to the output. The model generates predictions for multiple future time periods simultaneously, allowing the system to plan resource allocation across different time scales while maintaining a unified prediction framework.
Solution Approach 2:
The patent creates a universal prediction model that can generate predictions for multiple computing applications and time periods using the same underlying framework. This multi-functional model handles diverse prediction tasks (different applications, time scales, interaction types) through a single unified approach, improving robustness without proportionally increasing complexity.
Data Source
AI summary
The present disclosure relates to systems, non-transitory computer-readable media, and methods for predicting account interactions with computing applications. In particular, in one or more embodiments, the disclosed systems determine user account data associated with one or more computing applications for a user account. Additionally, in some embodiments, the disclosed systems generate, based on the user account data, a transition matrix and an emission matrix corresponding to a plurality of hidden states of a hidden Markov model to customize the hidden Markov model for the user account. Furthermore, in some implementations, the disclosed systems determine, utilizing the customized hidden Markov model, one or more predicted account interaction metrics for the user account in connection with the one or more computing applications based on the transition matrix and the emission matrix.