مبينا تاجي

Nowadays, one of the most popular and challenging tasks in natural language processing is answering complex user questions. Question-answering systems, as a new generation of search engines, receive user questions in natural language without semantic limitations and provide their answers in the closest and most comprehensible way possible for humans, i.e. in natural language. In recent years, most research in the field of question-answering systems has been focused on the English language, and not much effort has been made in languages with limited resources, such as Persian. This limitation prevents these systems from efficiently handling questions in languages like Persian. In this thesis, to solve the challenge raised in the Persian language and improve the efficiency of Question-answering systems in answering questions, the goal was to produce a correct and valid dataset for complex multi-step questions supported by these systems. Therefore, the first open-domain question-answering dataset in Persian multi-hop questions has been developed based on the unstructured Persian Wikipedia encyclopedia containing 7000 multi-hop questions. This dataset has been eva‎luated and fine-tuned on the latest pre-trained language models that support the Persian language, including ParsBERT, AlBERT, and RoBERTa; and the results of this eva‎luation are reported based on the two criteria of reliability and exact matching. The best results obtained based on the two criteria of reliability and exact matching on this dataset are 42.85 and 69.83, respectively. The results indicate that this dataset is a powerful starting point for improving multi-hop complex question answering for Persian question-answering systems.

سامانه‌هاي پرسش و پاسخ , پرسش چندگامي , دامنه باز , بهبود كارايي

Question Answering System , Multi-hop Question , open domain , Effectiveness Optimization

Mobina Taji

Dr. Hassan Naderi