How to determine relationship between two entities when there is more than one relation while creating distant supervision training data?
I got the concepts of distant supervision. As for my understanding, the creating training data process is like;
- Extract named entities from sentences
- Find two entities named "e1" and "e2" from each sentence.
- Search these two entities in knowledge base (freebase etc.) to find relationship between them
I got confused at this step. What if there is more than 1 relation between these two entities (e1 and e2) ? If so which relation should I select?
Solution
It depends on the model you're training.
Are you learning a model for one kind of relationship and doing bootstrapping? Then only pay attention to that one relationship and drop the others from your DB.
Are you trying to learn a bunch of relationships? Then use the presence or absence of each as a feature in your model. This is how Universals Schemas work.
Here's an image of a feature matrix from the Universal Schema paper:
- How to save checkpoints for thie transformer gpt2 to continue training?
- Bert model splits words by its own
- Searching for specific words in Corpus with R (tm package)
- Avoiding overlap in frequency and document frequency count in Quanteda
- ImportError: cannot import name 'deprecated' from 'typing_extensions'
- Questions about training LLMs on large text datasets for text generation from scratch
- HuggingFace Transformers For Text Generation with CTRL with Google Colab's free GPU
- Counting the Frequency of Some Words within some other Key Words in Text
- How to get Bigram/Trigram of word from prelisted unigram from a document corpus / dataframe column
- Error in getting Captum text explanations for text classification
- euclidian distance from word to sentence after doing Vectorizer
- Llama-3.2-1B-Instruct generate inconsistent output
- llama-cpp-python not using NVIDIA GPU CUDA
- Normalization of token embeddings in BERT encoder blocks
- How to recognize if string is human name?
- NER versus LLM to extract name, gender, role and company from text
- Is there a way to extract information from contracts using ML with including contract files and targeted strings as inputs and outputs?
- Haystack: save InMemoryDocumentStore and load it in retriever later to save embedding generation time
- Using an AWS service to execute a python script that will extract keywords from text using keyBERT?
- Hugging face tokenizer cannot load files properly
- How to convert character indices to BERT token indices
- How can I share a complex spaCy NLP model across multiple Python processes to minimize memory usage?
- Finding Root Form of Verbs using Curiosity-AI/Catalyst
- Is it possible to get embeddings from NV-Embed using Candle?
- Dutch sentiment analysis RobBERTje outputs just positive/negative labels, netural label is missing
- Target modules for applying PEFT / LoRA on different models
- How to reconstruct text entities with Hugging Face's transformers pipelines without IOB tags?
- How to guess, with spaCy, the frequency of a lemma or a word form in general language use?
- Importing deepspeech error: Module not found
- Question about data_collator throwing a key error in Hugging face