This project shows the usage of hugging face framework to answer questions using a deep learning model for NLP called BERT. This work can be adopted and used in many application in NLP like smart assistant or chat-bot or smart information center.
To search for an answer to a question from a PDF, use the searchAnswerPDF.py code.
To search for an answer to a question from a text, use the searchAnswerText.py code.
| NOTE: Running the code requires a proper internet connection for downloading the model from huggingface or should manually download all the files into a folder named bert-large-uncased-whole-word-masking-finetuned-squad and save the folder in the working directory. Files can be found in the following link |
| INSTALLATIONS: Please have the following libraries installed torch, transformers, PyPDF2 |
To use the searchAnswerPDF.py, the following parameters have to be tweeked as per your application.
# The question that you want to ask
question = 'What is life expectancy of kompsat-3?'
# The full path of the PDF from which you choose to take the context from
pdf_path='D:\\Projects\\bhoonidhi\\kompsat.pdf'
# The parent path of the working directory where the folder containing model files is present
model_path='D:\\Projects\\bhoonidhi\\'
To use the searchAnswerText.py, the following parameters have to be tweeked as per your application.
# The question that you want to ask
question = 'What is life expectancy of kompsat-3?'
# The text that you wish to take as context
text = '''KOMPSAT-3 is a high performance remote sensing satellite, which provides 0.7 m GSD
panchromatic image and 2.8 m GSD multi-spectral image data for various applications.
KOMPSAT-3 was launched into a sun synchronous low Earth orbit on the 18th of May, 2012
and the life time of more than 7 years is expected.'''
# The parent path of the working directory where the folder containing model files is present
model_path='D:\\Projects\\bhoonidhi\\'
The base code for this project is explained in detail below:
from transformers import BertForQuestionAnswering
from transformers import BertTokenizer
The above lines imports the model itself and the tokenizer algorithm
model = BertForQuestionAnswering.from_pretrained('bert-large-uncased-whole-word-masking-finetuned-squad')
tokenizer = BertTokenizer.from_pretrained('bert-large-uncased-whole-word-masking-finetuned-squad')
question = 'When was kompsat-3 launched?'
text = '''KOMPSAT-3 is a high performance remote sensing satellite, which provides 0.7 m GSD
panchromatic image and 2.8 m GSD multi-spectral image data for various applications.
KOMPSAT-3 was launched into a sun synchronous low Earth orbit on the 18th of May, 2012
and the life time of more than 7 years is expected.'''
input_ids = tokenizer.encode(question, text)
print("The input has a total of {} tokens.".format(len(input_ids)))
tokens = tokenizer.convert_ids_to_tokens(input_ids)
for token, id in zip(tokens, input_ids):
print('{:8}{:8,}'.format(token,id))
The tokenizer encodes the text into numbers using encode function.
output = model(torch.tensor([input_ids]), token_type_ids=torch.tensor([segment_ids]))
The encoded text is sent to the model as a torch tensor
answer_start = torch.argmax(output.start_logits)
answer_end = torch.argmax(output.end_logits)
if answer_end >= answer_start:
answer = " ".join(tokens[answer_start:answer_end+1])
else:
print("I am unable to find the answer to this question. Can you please ask another question?")
print("\nQuestion:\n{}".format(question.capitalize()))
print("\nAnswer:\n{}.".format(answer.capitalize()))
Below is an example of the model when asked a question based on the context provided.
Context:
KOMPSAT-3 is a high performance remote sensing satellite, which provides 0.7 m GSD
panchromatic image and 2.8 m GSD multi-spectral image data for various applications.
KOMPSAT-3 was launched into a sun synchronous low Earth orbit on the 18th of May, 2012
and the life time of more than 7 years is expected.
Question:
What is life expectancy of kompsat-3?
Answer:
More than 7 years.