Movies and Pop Songs Recommendation System by Emotion Detection through Facial Recognition
Jing‐ye Zhang · Journal of Physics Conference Series · 2020
Abstract In this paper, a convenient, accurate, and widely applicable recommendation system of films and pop songs, using deep learning as an efficacious tool, is propounded. We devised a refined Deep Residual Network (ResNet-38) for the emotion detection of the users, which achieves an accuracy of 64.02% for the testing set of Kaggle-fer2013. Other traditional methods including the use of Classifier SVM or four-layer CNN produce the average accuracies of 40.7% and 60.7% respectively. Thus, it is cogent to conclude that our model outperform other traditional models. The usual emotion-based movies or songs Recommendation systems include web scrawling of the user’s personal information, like his or her recent comments, based on which researchers construct “interest models” to achieve an understanding of the user’s emotion. However, this type of method, though comprehensive, infringes the users’ privacy, presents biased results, and lacks instantaneity and interactivity. Thus, this research paper serves to introduce a novel mode of Recommendation system so as to counter theses drawbacks.