Misspelling Detection by using Multiple Bidirectional LSTM Networks

Abstract

Companies in the RECRUIT Group provide matching business between clients and customers, and create lots of manuscripts every day in order to tell the attractiveness of our clients. In this paper, we propose a method for detecting misspelling in manuscripts with machine learning. That system mainly consists of two parts. One is the multiple Bidirectional LSTM networks to estimate the probabilities of correctness in each characters. The other is the random forests algorithm to decide what sentence is correct or not by using outputs of these networks. The efficacy of our approach is demonstrated on two datasets: artificial sentences and real manuscripts created in our services.