Word recovery in word-lost documents using an artificial intelligence methodology

Abstract

Documents are often physically damaged in part. Even if you save a damaged document online, it will be hard to recover it. Therefore, this study aims to develop an artificial intelligence methodology that can guess and recover lost words in documents. This study demonstrates that the Skip-Gram model can be used to construct relationships between words using the word embedding technique, a simplified form of the neural network language model. It also shows that the relationship between words can be used to infer words to fit in a specific part of a lost document.