Abstract:

In telecom communication network,a hot customer complaint often affects hundreds even thousands of users’ service and leads to significant economic losses and bulk complaints.An approach was proposed to predict a customer complaint based on real-time user signaling data.Through analyzing the network business layer logic,30 key segments related to the user experience in the S1 interface data were selected.Further,one-hot features,statistical derived features,and differential features were extracted to classify user perceptions in detail.Considering the problems of noise data and unbalanced training samples,LightGBM was chosen to train the prediction model.Experiments are conducted to prove the effectiveness and efficiency of this proposal.As of today,this approach has been deployed in our daily business to locate the hot complaint problem scope as well as to report affected users and area.

Key words: prediction of the complaint hotspot problem, signaling data, feature extraction, LightGBM classifier