<scp>AVQANet</scp> : Amharic Visual Question and Answering Model Based on Deep Learning Approach for Ethiopian Museum Visitors

ABSTRACT This study aims to design an effective deep learning‐based Amharic Visual Question Answering (AVQA) model for Ethiopian museum visitors by identifying a suitable algorithm architecture. The linguistic complexity of the Amharic language, coupled with challenges in artifact recognition and the lack of intelligent museum guidance systems, has limited access to Ethiopia's cultural heritage knowledge. To address these challenges, we proposed AVQANet, a model that processes artifact images and Amharic natural language questions to generate precise, context‐aware answers, thereby enhancing museum interaction, cultural preservation, and tourism experiences. The Convolutional neural network (CNN) used for image feature extraction and (LSTM, BiLSTM, GRU, and BiGRU) algorithms are used for text feature extraction within element‐wise combination mechanisms to generate answers. The dataset was collected from the National Museum of Ethiopia (NME) and Lake Tana monasteries museum by capturing slit lamp camera. The Keras python package with the backend of TensorFlow used for implementation setup. The findings showed that AVQANet model achieved a CNN and BiGRU based on Amharic visual question and answering classification result of 80.48% training accuracy and 75.60% testing accuracy. Unlike most existing VQA systems, which are primarily designed for English and rely on large‐scale open‐domain datasets, our study introduces the first AVQANet model tailored for Amharic language in Ethiopian museum artifacts. The AVQANet model is found to be faster to train and has a smaller model size compared to state‐of‐the‐art models.