| Abstract : | Natural Language Processing (NLP) and Machine Learning (ML) concepts can serve as a powerful tool in an era of rapidly growing data. It can reveal unseen information from the vast amounts of unstructured textual data, which can be helpful in adjusting business strategy. The primary purpose of this graduation thesis Text Analytics –Identifying contact reason based on the call content is to determine whether it is feasible to create insights from automatically transcribed calls and identify the reason behind customers contact. In the age of automation and efficiency it is necessary to search for new ways to gain actionable data and fully explore its value. To successfully execute the study, over 11,000 transcripts were used from calls across various call center domains. The proposed approach classifies the calls according to its content. After data preparation and feature extraction using TF-IDF text vectorization, calls were separated into tree clusters, reflecting main call categories in the given dataset: vehicles, parking pass and general inquiries. Upon the selection of calls from category vehicles Topic Modeling was performed on analyzed data using Latent Dirichlet Allocation (LDA) based model. Topic Modeling results identified six topics within selected cluster. The output of LDA model was a list of words and associated probabilities which were used to interpret and label topics of the model. Assigned topic labels reflected main inquiries regarding vehicles: questions about purchasing, discussing a deal, application for a credit, issues with warranty, exchanging of contact details and searching for information on the website.Each of above-mentioned topics could be assigned with different suggestions and action plan. Such insights could assist businesses to take data driven decisions and efficiently allocate their resources. Further studies are needed to take place in order to explore to the underlying patterns within the topics and call characteristics.
|
|---|
Copyright © 2013 Library AUEB. All rights reserved.