Neural Network Trigger Function for Dynamic Event Influence Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional Hawkes process models assume constant influence of past events over time, leading to reduced accuracy in predicting events like social media propagation and infectious disease spread, as the mode of propagation changes with time based on human activity levels.
Innovation Solution
A learning method that trains a point process model with an intensity function using a neural network to represent temporal changes in influence, where the trigger function is a composite function of a first function and its derivative, allowing the model to account for time-dependent propagation modes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If the trigger function is assumed to be constant regardless of absolute time point, then the model structure is simple, but the prediction accuracy deteriorates because the influence of past events does not reflect real-world temporal variations
Solution Approach 1:
The patent transforms the static trigger function into a dynamic one by introducing time-dependent parameters. The trigger function h(Δt|θ) is extended to h(Δt, t|θ) where t represents absolute time, allowing the influence of past events to vary dynamically with the current time point. This enables the model to capture temporal variations in propagation modes while maintaining a relatively compact mathematical structure.
Solution Approach 2:
The patent changes the parameters of the trigger function to include absolute time information. By modifying the parameter set θ to include time-dependent components, the model can adapt the influence decay characteristics according to the time of day or other temporal patterns, thereby improving prediction accuracy without fundamentally changing the model architecture.
2Measurement precision
If the trigger function incorporates temporal changes in influence, then the prediction accuracy improves, but the model complexity increases
Solution Approach 1:
The patent introduces dynamic elements into the trigger function by making it dependent on absolute time t in addition to the time difference Δt. This dynamic extension allows the model to capture temporal patterns such as diurnal variations in social media activity while maintaining the fundamental point process framework, thus improving accuracy without excessive complexity.
Solution Approach 2:
The patent modifies the parameter set of the trigger function to include time-dependent parameters. By extending θ to capture temporal variations, the model can adapt to changing propagation modes throughout the day without requiring a complete restructuring of the model, balancing accuracy improvement with controlled complexity increase.
Data Source
AI summary
A learning method to be executed by a computer according to one embodiment includes of acquiring event history information representing a history of a predetermined event; and training, by using the acquired event history information, parameters of an intensity function in which a trigger function is set to be a function represented by a composite function of a first function and a predetermined second function; and a derivative of the first function, the first function being represented by a neural network that models a temporal change in influence of the event.