Hello to everyone. I want to build a chatbot for my students in order to answer mathematical questions for the greek language. What I want to use is a question answering bert model or a sentence pair similarity. After trying various multilingual models pretrained on the question answering task on close domain i didnt have any luck, mainly because the text has specific mathematical terminology, complementary angles, supplementary angles e.t.c.
I have found a greek language bert model nlpaueb/bert-base-greek-uncased-v1 · Hugging Face which was trained on the greek part of wikipedia. Should I use this model and fine tune it on the greek part of Wikipedia on articles containing mathematical text and the training that model for the question answering task? And if this is the case does anyone know any question answering dataset for the greek language like the squad dataset? If my understanding is correct auto translate the squad dataset won’t give good results since after translation tha starting point of the answer may have changed. I would appreciate if someone could give me some quide lines to follow.