Abstract:
"IT BPO organizations rely heavily on customer satisfaction for customer retention. When a
customer is being serviced through these organizations, a key communication method is
through voice calls. Understanding the customer sentiment through these voice calls, could
provide a good indication of the customer satisfaction level, thus allowing the organization
to have superior customer retention and gain a competitive advantage. This solution aims to
provide an efficient process to analyze these voice calls using machine learning and NLP
techniques. The automated solution involves transcribing the voice call to text, then
performing classification using machine learning to gauge the sentiment. This data can be
taken as a KPI to improve overall customer satisfaction.
A suitable dataset was selected for the training, seven traditional classification models and
2 deep learning models were selected for training. The models were evaluated using key
evaluation metrics. Out of these, the logistic regression model outperformed the rest of the
models. From the deep learning models, LSTM scored the highest accuracy. For the user
interface, the logistic regression and LSTM models were used to gauge the sentiment of the
uploaded audio file."
