Bursty Events Detection with the Field of Mobile Customer Service

Abstract

In the field of mobile customer service, the increase of traffic volume and the drop of connection rate caused by uncertain factors are called bursty events. When bursty events occur, detecting the bursty events timely and proactively can improve resource scheduling efficiency, connection rate, and customer satisfaction. The existing bursty events detection methods are mainly dependent on human experience, which detect events untimely and incompletely. In this paper, an unsupervised approach of detecting bursty events based on speech-to-text data is proposed, which makes good use of multiple dimensional features of the field to detect and track bursty events. Using our method, we achieve performances of 90.46%, 86.22% and 86.15% w.r.t. the average precision, recall and F1 score respectively. The experimental results demonstrate that the proposed method is effective to detect bursty events among considerable speech-to-text data.