An application for detecting network related problems from call center text data

A Network Service Provider can receive different network related problems and complaints from various communication channels regarding their service activity in certain regions. One of the major and most important communication channels is customer call center. Detecting network related problems that customers are notifying during these calls are significant in order to provide solutions and increase customer satisfaction. However, due to sheer volume of the call records that are converted to text, it is quite difficult to analyze whole data using traditional approaches. In this paper, we study a topic modeling approach for detecting network related problems from call center text data. The analysis results demonstrate that for a major broadband service providers' personal Internet at home tariff, most of calls received in a customer call center is related to information, whereas the second majority of all calls are related to faults that are network related issues. These results signify the existence of network and service related issues in service providers' infrastructure.

I. Introduction

Today's service providers are receiving a growing number of data from various communication channels that they are interacting with their subscribers who are may evaluate their products and services. The data can arrive from different communication channels including social media, customer call center calls, emails, forums and blogs, customer reviews, surveys and chat [1]. This data, on the other hand, is a significant asset for gaining competitive advantages for service providers. It is mostly not easy to extract relevant information from this mostly unstructured data.