Object Linking Model Training via Confidence Difference Weak Supervision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The high manual costs and difficulty in labeling training datasets for object linking models in human-machine interaction tasks, such as semantic parsing and intelligent question answering, hinder the development of accurate and efficient models.
Innovation Solution
A weak supervision mechanism is introduced, where a target semantic object's confidence scores are determined with and without specific text elements, allowing the object linking model to be trained based on confidence differences, reducing the need for precise labeling and improving training efficiency and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large scale training dataset is prepared for object linking model, then the model training accuracy is improved, but the manual labeling cost and difficulty increase significantly
Solution Approach 1:
The system uses the object linking model itself to generate training data by automatically linking text elements to semantic objects without manual intervention. The model processes text sequences and generates confidence scores, using its own output to create training datasets, thereby eliminating the need for expensive manual labeling while maintaining training accuracy
Solution Approach 2:
A confidence score generation module is introduced as an intermediary between the text input and the training data output. This module automatically generates confidence scores indicating the likelihood of semantic object mentions, serving as a bridge that converts raw text into structured training data without requiring manual annotation
2Manufacturing precision
If manual labeling is performed to ensure high accuracy, then the training data quality is improved, but the productivity and efficiency decrease
Solution Approach 1:
The system automatically generates training data using the object linking model's own capabilities. The model processes text sequences, identifies semantic objects, and generates confidence scores automatically, enabling the system to produce high-quality training data at scale without manual intervention, thereby maintaining data quality while dramatically improving productivity
Solution Approach 2:
The system performs preliminary processing of text sequences by pre-extracting semantic objects and generating confidence scores before the actual model training. This preliminary action creates ready-to-use training data that can be directly fed into the training pipeline, eliminating the need for time-consuming manual labeling while ensuring data quality
Data Source
AI summary
According to implementations of the present disclosure, there is provided a solution for training an object linking model. A target semantic object and a first text sequence comprising text elements in a natural language are obtained. A first confidence score of the target semantic object being mentioned in the first text sequence is determined. A second confidence score of the target semantic object being mentioned in the first text sequence is determined with a first text element being ignored from the first text sequence. An object linking model is trained at least based on a first confidence difference between the first and second confidence scores, the first text sequence and the target semantic object. In this way, the cost and difficulty in labeling a training dataset may be reduced and the labeling accuracy and efficiency may be improved.