User Action Waveform Imaging for Time-Series ML Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies struggle to capture time-series changes in user action characteristics, such as execution frequency and cycle, as explanatory variables in machine learning, making it difficult to predict medium to long-term indicators like user satisfaction levels accurately.
Innovation Solution
An information processing device that generates waveform data from user action data, connecting time-series timings with predetermined waveforms and performs power spectrum imaging to create two-dimensional images representing action characteristics, which are used as explanatory variables in machine learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If user action data is input as a total number (statistical value) of targets within a certain period, then the machine learning model can process the data easily, but the model cannot capture time-series changes in action characteristics such as execution frequency and execution cycle
Solution Approach 1:
The patent transforms one-dimensional statistical values into two-dimensional time-series data by plotting action timing points on a timeline and connecting them with waveforms. This dimensional transformation enables the representation of temporal patterns, execution frequencies, and cycles that are lost in aggregate statistics, while the resulting images can still be processed by machine learning models.
Solution Approach 2:
The patent introduces waveform connections as an intermediary between raw action timing data and machine learning input. These waveforms serve as a mediator that encodes temporal relationships and action characteristics, bridging the gap between simple statistical data and the need for time-series information without requiring complex data processing pipelines.
2Measurement precision
If waveform data is generated by connecting time-series timings with predetermined waveforms, then time-series information such as execution frequency and execution cycle can be captured, but the complexity of data processing increases
Solution Approach 1:
The patent changes the representation parameters of action data from raw timing values to waveform-based visual representations. By transforming timing information into waveform patterns with specific frequencies and periods, the system captures action characteristics more precisely while the resulting images provide an intuitive representation that simplifies further analysis.
Solution Approach 2:
The patent creates a visual copy or representation of the time-series action data through waveform connections and power spectrum imaging. This visual copy preserves the essential temporal patterns and action characteristics while transforming the data into a format that is easier to interpret and process, effectively copying the information in a more manageable form.
3Loss of information
If power spectrum imaging is performed on waveform data, then a two-dimensional image representing action characteristics is generated for use in machine learning, but the processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary processing by generating waveform data and visual representations before feeding data into machine learning models. This preliminary action of creating waveform connections and power spectrum images allows the system to pre-extract and organize temporal patterns, reducing the computational burden during subsequent machine learning processing and enabling more efficient model training.
Data Source
AI summary
An information processing device (10) includes: a waveform data generation unit (12) that, from action data related to acquired action of a user and including a type and an execution time information of the action, obtains a timing in time series of an event related to action characteristics of the user and connects the timings adjacent in time series with a predetermined waveform, to generate waveform data for each action; and an image generation unit (13) that generates a two-dimensional image representing the action characteristics of the user by performing power spectrum imaging on the generated waveform data.