PerAnSel:  A  Novel Deep Neural Network-Based System for Persian Question Answering

Figure 4

In this figure, green components indicate the main components of PerAnSel model. Orange, red, and dark-green vectors show the output of [CLS] token, CNN, and RNN networks respectively. (1) The question is passed to the Question Processing and the EAT is detected. (2) In the information retrieval, the most relevant paragraphs to the question are retrieved from documents. (3) The question and its relevant paragraphs are transferred to the Answer Extraction to find the exact answer from each paragraph. (4) The EAT, the question, and extracted exact answers are passed to the Answer Selection, and (5) the most correct answer is returned to the user (violet).