Skip to content

A closed domain model for question-answering in Russian built with transfer learning techniques. The model is fine-tuned on a custom dataset collected with the methodology described in the SQuAD original paper.

License

Notifications You must be signed in to change notification settings

vifirsanova/nlp-huawei-project

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

49 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

A Russian Question Answering System for Inclusive Education

The dataset used for the model training: https://doi.org/10.6084/m9.figshare.13295831

The repository contains the code for the data preprocessing, model fine-tuning and evaluation of two transformers adapted for the question answering task:

  1. BERT based QA model
  2. GPT-2 based QA model

The BERT based system implementation is based on Question Answering with SQuAD 2.0 guide: https://huggingface.co/transformers/custom_datasets.html?highlight=custom#qa-squad

The GPT-2 based system implementation is based on by gpt-2-simple package by Max Woolf: https://minimaxir.com/2019/09/howto-gpt2/

Results

Base model F-Score
Multilingual BERT 0.55
774M GPT-2 0.63

Sample outputs

BERT based model

Q: Есть ли какие-то техники, чтобы научить ребенка с РАС вести себя в социальном мире?

A: Поговорите со специалистами, работающими с ребёнком, о каких-либо программах, которые могут ему помочь.

GPT-2 based model

Q: У детей-аутистов могут быть проблемы с одноклассниками?

A: У детей с расстройством аутистического спектра (РАС) высок риск подвергнуться травле, особенно в обычных школах.

About

A closed domain model for question-answering in Russian built with transfer learning techniques. The model is fine-tuned on a custom dataset collected with the methodology described in the SQuAD original paper.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published