This document describes how to train a model for text relation extraction.
After you build the dataset, you can begin training the model. Return to your project, switch to "Model Hub", and click "Create Model".
On the model creation page, the platform handles all implementation details. You only need to select a model to start training. First, enter a name for your model. For Model type, select "Relation Extraction PCNN", "Relation Extraction Bert", "Relation Extraction BertNoise", or "Relation Extraction StructBERT-split". For more information, see the "Model descriptions" chapter.
The "Relation Extraction PCNN" model is selected by default. Click the add training data button to select your annotated or uploaded datasets. Then, click Submit. The model automatically starts training.
After initialization, model training starts automatically. Click "View" to go to the model details page, where you can check the training logs and specific evaluation metrics.
The evaluation metrics for entity extraction include Precision, Recall, and F1 score. The value for each metric ranges from 0 to 1. Higher values indicate better model performance.
Precision: For a specific class, this is the ratio of correctly predicted samples to the total number of samples predicted for that class. For the entire model, this is the ratio of all correct predictions to the total number of predictions.
Recall: For a specific class, this is the ratio of correctly predicted samples to the total number of actual samples in that class. For the entire model, this is the ratio of all correct predictions to the total number of actual samples.
F1-score: The harmonic mean of precision and recall.