Paper Title
Design of Indonesian Document Classification Framework based on Multimodal Deep Networks

Document Filling is an important part of office administration. One of the filing tasks is the classification of archived records which is still done manually and repeatedly causing inefficient human resources, time, and costs. Most archival documents are scanned and photographed documents that are saved in image or pdf format. Many machine learning models have been developed to complete classification work by extracting visual features and text in documents. This study proposes a framework design for a specific problem, namely the classification of Indonesian archival documents based on multimodal deep networks by utilizing pre-trained models and Indonesian corpora. Experiments and evaluations show that the framework design can provide optimal and better classification accuracy according to the baseline model by using a pre-trained model that has been widely developed. Keywords - Document Image Classification, Multimodal, Transfer Learning, Software Engineering.