Sequence Mining Model for Tagged and Untagged Data Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing supervised learning solutions for sequence data in risk management scenarios are limited to scenarios with tagged sequence samples, lacking universality and leading to resource waste and reduced accuracy due to the inability to utilize untagged data.
Innovation Solution
A sequence mining frame with multiple sub-models is employed to construct a sequence mining model, allowing for the selection of appropriate sub-models based on tag status, enabling training with both tagged and untagged historical sequence data to improve resource utilization and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a supervised learning solution with DTW algorithm and KNN model is used to mine sequence data, then the accuracy of distinguishing good and bad sequence samples is improved, but the solution becomes inapplicable to scenarios without tagged sequence samples, reducing universality
Solution Approach 1:
The patent segments the sequence mining task into multiple sub-models (first sub-model for latent representation, second sub-model for tagged data, third sub-model for untagged data). Each sub-model handles specific types of sequence samples based on their tagging status, allowing the system to process both tagged and untagged data appropriately while maintaining high accuracy for each segment.
Solution Approach 2:
The patent creates a universal sequence mining model that can handle multiple scenarios through a unified architecture. The model accepts sequence samples regardless of tagging status and routes them to appropriate processing paths (second sub-model for tagged, third sub-model for untagged), making it applicable to both scenarios with and without tagged data while maintaining consistent performance.
2Measurement precision
If only tagged sequence samples are used for training the sequence mining model, then the model accuracy is improved, but the resource utilization decreases due to inability to use untagged historical data
Solution Approach 1:
The patent applies preliminary action by using the first sub-model to extract latent representations from all sequence samples (including untagged ones) before classification. This preprocessing step prepares the data in advance, allowing untagged samples to be effectively utilized in the third sub-model without compromising the accuracy that would normally require tagged samples.
Solution Approach 2:
The patent introduces an intermediary mechanism (the first sub-model generating latent representations) that bridges tagged and untagged data. This intermediary processing transforms untagged sequence samples into a format that can be effectively used for training, enabling the system to leverage all available historical data while maintaining model accuracy through the coordinated work of multiple sub-models.
Data Source
AI summary
The present disclosure includes a method for training a sequence mining model. In the method, a first sequence sample in a target service scenario is obtained. A tag status of the first sequence sample is obtained, the tag status of the first sequence sample indicating a proportion of the first sequence sample that has corresponding tag information. A sub-model from a sequence mining frame is selected according to the tag status to construct the sequence mining model. Also, the sequence mining model is trained by using the first sequence sample. The sequence mining frame includes a first sub-model configured to obtain a latent representation, a second sub-model configured to determine the target tag information when the tag status meets a first condition, and a third sub-model being configured to determine the target tag information when the tag status meets a second condition.