Dynamic Thematic Relationship Identification in Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning models for identifying thematic relationships are static and unable to adapt to evolving terminology and changing contexts over time, making them ineffective in dynamic data sets.
Innovation Solution
Training machine learning models using temporally relevant data to determine thematic relationships at specific time periods, allowing for the identification of changes in terminology, sentiment, and trends over time, and providing a user interface for querying and visualizing these changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a conventional static machine learning model is used to identify thematic relationships, then the system structure is simple and easy to implement, but the model cannot adapt to evolving terminology and changing contexts over time
Solution Approach 1:
The patent implements dynamic thematic relationship identification by training separate machine learning models for different time periods. Instead of using a single static model, the system dynamically adapts to changing terminology and contexts by selecting or retraining models with temporally relevant training data, allowing the system to evolve with the data while maintaining manageable complexity through modular model architecture.
Solution Approach 2:
The system changes the temporal parameters of the training data to match the query time period. By adjusting the time period parameter of the training dataset to correspond to the query time period, the model adapts its thematic relationships to reflect evolving terminology and contexts without requiring complete system redesign.
2Loss of information
If training data from a single time period is used, then the model training is simple and fast, but the system cannot identify trends or changes over time
Solution Approach 1:
The patent segments the training data into multiple time periods, with each segment used to train a specialized model for that period. This segmentation allows the system to identify trends by comparing results across different time periods while keeping individual model training efficient and focused on specific temporal characteristics.
Solution Approach 2:
The system performs preliminary training of multiple models across different time periods before actual querying. By pre-training models with temporally relevant data for various time periods in advance, the system prepares trend analysis capabilities without incurring training time during actual query operations, thus losing minimal time during production use.
3Measurement precision
If a static model is used, then the processing speed is fast and consistent, but the accuracy decreases when terminology or contexts change
Solution Approach 1:
The system dynamically selects or retrains models based on the temporal characteristics of the query data. When terminology or contexts change, the system adapts by using models trained on temporally relevant data, maintaining high accuracy without requiring complete retraining. This dynamic approach preserves processing speed through selective model application rather than universal retraining.
Solution Approach 2:
The system employs periodic retraining or model selection based on time periods. By periodically updating or selecting models that correspond to the current time period's characteristics, the system maintains accuracy with evolving terminology while avoiding continuous retraining, thus preserving processing efficiency through structured periodic updates.
Data Source
AI summary
Methods and systems are described herein for improvements to identifying thematic relationships in data sets using machine learning models. In particular, the methods and systems describe a way to identify dynamic thematic relationships (e.g., thematic relationships that may change as a function of time) as a function of time.