FarCQA: A Farsi Community Dataset for Question Classification and Answer Selection

Question Answering Systems (QASs) have become increasingly important due to the need for accurate and concise answers that traditional search engines often struggle to provide. However, the development of QASs for the Persian language has been limited due to its complexity, fewer available resources, and tools compared to other languages. One crucial component of a QAS is question classification, which plays an effective role in retrieving correct answers. In this paper, we introduce FarCQA, the first open domain Persian community dataset for question classification and answer selection tasks, collected from an online forum. This dataset is tagged with 9 types of questions and includes both formal and informal language. In addition, we propose question classification and answer selection models using transformer based models and combining word embedding and deep learning techniques. Our approach demonstrates a notable accuracy on the test set, surpassing state-of-the-art methods.