Maximizing the Voice of the Customer with NLP: A Tool to detect Insights with Sentiment Analysis and Named Entity Recognition

The use of machine learning techniques to analyze customer feedback has the potential to provide valuable insights for businesses. In this thesis, we present a tool for extracting Voice of the Customer (VoC) data from tweets and analyzing it using natural language processing (NLP) techniques. The tool consists of four macro steps that are flexible and can be run in parallel, allowing for efficient and effective data extraction and analysis. We propose two architectures one for Sentiment Analysis and one for Named Entity Recognition (NER), and demonstrate that our approach outperforms the baselines reported in the literature in both tasks. In the case of Sentiment Analysis, we address the data and conceptual shift after demonstrating it with the Human Level Performance Analysis. For NER, we propose a method for processing the data to leverage different word embeddings and a way to conduct ablation studies using TensorBoard. The tool has been deployed and is being used by Toyota Motor Europe to conduct real-time VoC analysis. The results of research show that the integration of NLP and VoC analysis can provide valuable insights for businesses and contribute to the field of machine learning.