NLP-Based Solutions for Call Center Optimization
Svetlana Segărceanu, Maria Niculae, Theodor Pintilie, George Dan Suciu, Inge Gavăt, Romulus Chevereșan, Gabriel Stoica, Marian Ceaparu, Andrei Dănilă · 2024
Call center data flow processes audio recordings and text messages exchanged between clients and operators. Speech-to-text technology transcribes audio recordings into text. Modern call centers use Natural Language Processing (NLP) to process text messages and transcriptions, extracting keywords, named entities, or summarizing texts. This enables the system to update customer profiles, outline products, or identify frequently asked questions. The paper presents some results of processing audio data from a Romanian call center. We developed a subsystem to transcribe spoken customer messages and extract keywords and named entities. To build customer profiles, we extracted information such as names, phone numbers, address elements, and purchased product details. For product definitions, we considered names and, if available, product codes. We used Python, Google Speech-ToText API, and libraries like Spacy, Rake, T5, and scikit-learn. Despite challenges, we drew valuable conclusions.