سيستم پرسش‌وپاسخ واقع‌نما از ترجمه فارسي قرآن كريم

Closed-Domain Question Answering , Quran’s Translation , Factoid Question , Language Model , Semantic Retrieva‎l

A Factoid Question Answering System on Persian Translation of Holy Quran

مهندسي نرم افزار

The Holy Quran is the most important source of knowledge in Islam. It is one of the main references that Muslims refer to and rely on to find answers to their religious questions. Nowadays, with the widespread use of the internet and significant advancements of Artificial Intelligence in the field of chatbots and Question Answering (QA) systems creation, users tend to get the answers to different types of questions in the shortest possible time. However, religious textual sources such as Quran, have some unique features that make it challenging for common QA systems to find the answer of Quranic questions which is obviously because these systems are trained with different types of textual resources. The same problem also exists for Persian translations of the Holy Quran. Quranic translations have a unique style and are different from other Persian textual resources in terms of structure and vocabulary. Furthermore, most of the existing Quranic sytems use traditional approaches which makes them inefficient to answer users’ questions. In this study, an attempt has been made to provide a closed-domain QA system in order to answer factoid questions from the Persian translation of the holy Quran. Due to the lack of a proper dataset, a dataset of question-verse-answer triplets has been created. The approach used for creating the dataset is a hybrid approach which is consisted of automatic, semi-automatic, and manual techniques. The obtained dataset has been used to train different components of the mentioned QA system. The proposed system comprises two main modules which are responsible for retrieving related verses and extracting the final answer from related verses respectively. language models have been used in the structure of both the retriever and reader modules. The QA system presented in this research achieved a recall of 39% and an F1 score of 85%. These results show the acceptable performance of the proposed system in retrieving related verses and finding accurate answers to factoid question from the Persian translation of the Quran. According to the eva‎luation results, the proposed system provides more accurate answers to qurainc questions compared to widely used LLM based models such as ChatGPT.

6