Abstract
Indian railways are one of the largest rail networks in the world, and millions of passengers travel daily through it, due to which there are also a vast number of complaints in front of Indian Railways coming every minute through various mediums like COMS (Complaint Management System) app, RailMadad, SMS etc. Given the top-down approach which is followed for the uncategorised complaints making official's work time-consuming. Therefore, faster complaint redressal becomes a critical factor for the passenger's satisfaction. Previous research has focused on traditional machine learning algorithms and Twitter dataset available publicly to tackle this problem. In this work, we have explored deep learning techniques on an official dataset of the COMS app from CRIS (Centre for Railways Information Systems) and proposed RailNeural: an Attention Based Bi-Directional Long Short-Term Memory (LSTM) model which analyses user's complaint input sequences, capturing the underlying character level feature and then classifies them into their respective departments of field units ensuring prompt and accurate redressal of complaints. Our model outperforms several baseline models achieving an accuracy of 93.25 per cent and an F1-Score of 0.93.