Question Answering On SQUAD

Short-Video

chatbot_record.mp4

Requirements

• Tensorflow=2.4.0
• CUDA=11.2
• CUDNN=11.4 (necessary only for gpu computing (optional))
• Python 3.6

Dataset

The Stanford Question Answering Dataset (SQuAD), which is derived from Wikipedia, can be used for question answering chatbot. The SQuAD includes:

• 107,785 question-answer pairs depend on 536 articles.
• Due to a lack of RAM, only 10.000 pairs have been used for training of the Seq2Seq model.
• According to the results, given questions to the model can be predicted by the model accurately.
• Given data to the model should be enhanced in order to increase the accuracy of the model.

Folders:

Data:

• Downloaded data will be saved in this folder. Dowloading script is available in "Preporeccsing file.py". The program will download it in Data folder, if the folder exists.

Weight:

• Model weight, encoder and decoder model will be saved in this folder.

Image:

• The history of the model will be saved in this folder.

Model:

Seq2Seq architecture has been chosen to train the model.

Running:

• The "main.py" file must be run in a virtual environment with the system requirements.