Natural Language Processing for Detecting Brand Hate Speech

Abstract

Brand hate is a complex feeling that is not easy for companies to recognize. Mednini and Turki (2022) have confirmed that the hate can originate from genuine brand haters or an employee who works with competitors, to spread negative word-of-mouth in communities. That is why it is important to detect this emotion. This study aims to identify brand hate speech based on NLP techniques to detect consumer hate sentiment using a chatbot. We present a methodology for fine-tuning the GPT 2 language model for sentiment analysis through text classification. Experiments are conducted on datasets in three languages — Arabic, French, and English — within the context of consumer consumption. The model is retrained on labelled data to effectively identify brand hate sentiment. Furthermore, we evaluate our chatbot by conducting semi-structured interviews with diverse consumers. The experimental results demonstrate a significant improvement in sentiment analysis performance, highlighting increased accuracy when compared to other models and baseline approaches. We achieved an accuracy rate of 0.98 in the training set and 0.84 in the testing set, showcasing the utility of using GPT-2 in this context. This research contributes to the capability of managers to promptly identify brand hate speech, and proactively avert potential brand crises.