A Text Classification Methodology to Assist a Large Technical Support System

Abstract

Text-based tools for reporting technical issues and receiving support are widespread in commercial applications, such as customer services and internal corporate communication. Past issues recorded in such systems may provide valuable knowledge for better handling future interactions. Nevertheless, the predominance of short messages and the presence of specific domain subjects constitute additional challenges. In this work, we aim to build an assistant for a system operating in a large company that provides asynchronous services for technical support. It is known that some repetitive technical issues can be handled with simple standard messages, named <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">templates</i> . Thus, we propose a modular pipeline based on natural language processing and machine learning algorithms to enable raw text processing, feature extraction, and supervised learning to recommend suitable templates from a given textual description of the incoming issue. In a real-world scenario, the proposed pipeline achieved an average accuracy of 72.7%, a promising result for a setup with 9 classes and few labeled training instances. Moreover, a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">post hoc</i> analysis shows how our methodology is able to correctly identify the words more closely related to the corresponding templates.