Automated hierarchical classification of scanned documents using convolutional neural network and regular expression

Abstract

<p>This research proposed automated hierarchical classification of scanned documents with characteristics content that have unstructured text and special patterns (specific and short strings) using convolutional neural network (CNN) and regular expression method (REM). The research data using digital correspondence documents with format PDF images from pusat data teknologi dan informasi (technology and information data center). The document hierarchy covers type of letter, type of manuscript letter, origin of letter and subject of letter. The research method consists of preprocessing, classification, and storage to database. Preprocessing covers extraction using Tesseract optical character recognition (OCR) and formation of word document vector with Word2Vec. Hierarchical classification uses CNN to classify 5 types of letters and regular expression to classify 4 types of manuscript letter, 15 origins of letter and 25 subjects of letter. The classified documents are stored in the Hive database in Hadoop big data architecture. The amount of data used is 5200 documents, consisting of 4000 for training, 1000 for testing and 200 for classification prediction documents. The trial result of 200 new documents is 188 documents correctly classified and 12 documents incorrectly classified. The accuracy of automated hierarchical classification is 94%. Next, the search of classified scanned documents based on content can be developed.</p>

Advanced content-based retrieval for digital correspondence documents with ontology classification

“…The pre-processing stage involves the development of the automatic hierarchical classification model of digital correspondence documents based on the text classification approach in line with [29] that have been successfully classified digital correspondence documents into 4 manuscripts of letter, 5 type of letter, 15 origins of letter, and 25 subjects of letter with 94% accuracy by combining CNN and regular expression methods. The automatic hierarchical classification model developed in this research covers all the criteria according to the hierarchy of digital correspondence documents which include 5 manuscripts of letters, 22 types of letters, 15 origins of letters, and 25 subjects of letters.…”

“…The automatic hierarchical classification model developed in this research covers all the criteria according to the hierarchy of digital correspondence documents which include 5 manuscripts of letters, 22 types of letters, 15 origins of letters, and 25 subjects of letters. The manuscript, origin, and subject were classified using the same regular expression pattern applied in the previous research [29] while the types of letters were based on the CNN. Moreover, this model was further applied in the archiving stage.…”

“…MAXENT function, SOFTMAX activation function, input (3 * cnnLayerFeatureMaps), and 22 output. It is important to note that there are 22 classifications based on the types of letters while the manuscripts have 5 classifications which were used as the hierarchical tree of the types of letters in line with the modified Regex pattern adapted from [29].…”

“…However, it cannot be used to classify different types of documents with unstructured text content. Another research by [29] automatically applied a multilevel classification method to scanned documents in image format based on the document hierarchy. OCR technology was utilized to recognize the text in these scanned documents in the form of digital correspondence documents and also to automatically classify the letters based on the document hierarchy up to a depth of 4 levels which include 4 classifications of the manuscripts of letters -> 5 classifications of the types of letters -> 15 classifications of the origins of letters -> 25 classifications of the subjects of letters by combining the CNN and regular expression method.…”

“…The overall review of previous research showed that content-based search has been widely applied to scanned documents using OCR but there is no specific focus on the classification as well as ontology construction as [32]- [34] to describe the hierarchical information on the scanned documents classified. Therefore, this present research is a significant improvement on [29] by developing classification capabilities for 5 to 22 types of documents. It also proposed a method to search classified content-based digital correspondence documents with automatic ontology relations to overcome problems associated with their searching process and search classified content by displaying ontological relationships to make it possible to easily trace the documents containing certain content and attribute information on the official manuscript, types, origin, and subject of the letters classified.…”