Triple Linking

Pipeline

1. Before conduct the unsupervised triple linking, you need to prepare a bert-like pre-trained model (BERT, SimCSE, BGE) as the textual encoder and change the base_model variable in the scripts.
2. Run the scripts triple_bge.py and question_bge.py to encode the triples and questions using semantic embeddings and run the script retrieve_relative_triple.py to retrive the top-K relative knowledge for each question.

The format of the question data in question_bge.py is the same as the data sample in data/ and we provide a KG data sample for reference.