Deep Learning-Based Facial Emotion Recognition for Detecting Brand Hate

This paper aims to recognise three levels of consumer hate – cold hate, cool hate and hot hate – through facial emotion recognition. It compares the use of transfer learning approaches and custom convolutional neural network (CNN) approaches. Five classes of emotions were considered in this research: fear, anger, disgust and contempt as indicators of a brand with hate, and happy as an indicator of non-brand hate. The databases used are AffectNet and RAF_DB. Multiple facial expressions are included in our dataset, which consists of about 23,529 images belonging to five classes in the training set and 10,088 images belonging to five classes in the testing set. The custom CNN model achieved an accuracy of 78.4% when validated on the testing set, compared to the VGG-16 model, which achieved an accuracy 79%, respectively. The confusion matrix was used to verify the findings, confirming that the custom CNN model outperforms the pre-trained models. Notably, it successfully detected brand hate emotions with an accuracy of 85%. This paper employs facial emotion recognition – a behavioural expression measure – to help companies capture the emotional responses of haters.