© unsplash/@machec
Classification of customer requests
Context
Customer satisfaction depends strongly on a fast processing of customer support requests. Since the customer requests can fall into a wide range of areas, requests need to be classified and assigned to the right customer agent or team in a short time.
Challenges
The classification of customer support requests can be automated by using a machine learning (ML) model, that extracts the relevant information, such as order number, claimed product and its description of requests which is often free text. To do this, text passages need to be abstracted from the requests and classified into customer service categories or tickets. For this, the model has to recognize semantically similar words in order to find the related customer service for the information provided by the customer request.
The biggest challenge is the analysis of the free text as customers expect the agent to know about previous communication and the current processing status.
Potential solution approaches
Extracting relevant information from text data is accomplished by using Natural Language Processing (NLP) techniques such as naive Bayes classifiers, TF-IDF algorithms or LSTM networks, which are able to recognize the relations between words and their respective context.
The basis for the training is a large data set, where the text passages relevant for the classification are labeled, so that the model is able to make classifications based on the extracted information.
Additionally, the current workflow in processing customer requests should be modelled and mapped as good as possible not to change already established workflows in the company.