Temporal Action Series Learning for Before-and-After Event Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing natural language processing technologies fail to effectively capture changes in user actions before and after an event due to a lack of consideration for the temporal context and series changes, limiting the interpretability of semantic vector closeness.
Innovation Solution
A learning device and method that utilize the Skip-Gram and CBOW neural networks to create models for estimating peripheral series and specific series changes in user actions by incorporating date and time labels, allowing for the conversion of discrimination labels to predict actions before and after an event.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If semantic vector closeness is used to extract actions, then the extraction can be performed mathematically, but the interpretability of changes before and after events is insufficient
Solution Approach 1:
The patent segments the action series into peripheral series (actions before and after the event) and specific series (the event itself), allowing separate analysis of temporal patterns. This segmentation enables precise detection of action changes while maintaining interpretability through structured temporal decomposition.
Solution Approach 2:
The patent introduces temporal dimension (before/after event discrimination labels) to the semantic vector space, transforming the problem from static semantic closeness to dynamic temporal pattern recognition. This dimensional expansion enables both precise action change detection and improved interpretability through temporal context.
2Loss of information
If position in sentence is considered for word representation, then semantic closeness can be expressed, but temporal context and series changes are not captured
Solution Approach 1:
The patent performs preliminary action by creating discrimination labels that categorize actions as before or after the event before feeding data into the neural network. This pre-processing of temporal information allows the model to capture temporal patterns without significantly increasing model complexity.
Solution Approach 2:
The patent uses discrimination labels as an intermediary between the raw action series and the neural network model. These labels encode temporal context information, serving as a bridge that enables the model to learn temporal patterns while keeping the underlying architecture relatively simple.
3Productivity
If only semantic vectors are used for action extraction, then the extraction process is simple, but the ability to identify specific action changes is limited
Solution Approach 1:
The patent merges semantic vector representation with temporal discrimination labels, combining both approaches to achieve both efficiency and precision. The neural network processes both semantic information and temporal context simultaneously, enabling accurate identification of specific action changes while maintaining computational efficiency.
Data Source
AI summary
Provided is a learning device 1 including: a first learning unit 103 that learns a first model 3 that estimates a peripheral series of a series from the series including one or more items by using training time series data including a plurality of series indicating an action of a user, each item in the training time series data being assigned with a date and time information label indicating a date and time when an action is performed and a discrimination label for discriminating before and after occurrence of an event; and a second learning unit 104 that learns a second model 4 that estimates a specific series in which a peripheral series exists in the periphery from the peripheral series by using the training time series data.