Create README.md
Browse filesA Q/A dataset from https://islamqa.info
Contains 82086 question and aswer pairs.
We provide in two separate files questions and answer embeddings.
In the questions file we provide metadata like categories, language, similar questions, etc.
In the embeddings file, we provide an embedding using bert-base-multilingual-uncased.
<code>
encoded_input = tokenizer_mutlti(answer, return_tensors='pt', truncation=True, padding='max_length', max_length=512)
with torch.no_grad():
output = model(**encoded_input)
last_hidden_states = output.last_hidden_state
last_hidden_states=torch.mean(output.last_hidden_state, dim=1)
</code>
We provide the answer embedding in addition to embeddings of every sentence in the answer. Sentences are identified using spacy xx_ent_wiki_sm model.