Using DNN Algorithm for Emotional Detection Based on Video Comments Text Data Herbert Siregar (a*), Erna Piantari (b), I Gusti Ngurah Agung A Ananta Wijaya(a)
a) Computer Science, Universitas Pendidikan Indonesia, Jalan Setiabudi 229, Bandung, Indonesia
b) Computer Science Education, Universitas Pendidikan Indonesia, Jalan Setiabudi 229, Bandung, Indonesia
Abstract
Youtube is the largest online video provider site on the internet which provides facilities for uploading various content according to applicable regulations. The video certainly impacts viewers, which can be seen in the comments that appear. Comments given by viewers can implicitly show the emotions they feel. Today, the extraction of emotional levels in comments is a part that is widely studied to see the impact of media on audiences. The benefits of this extraction can be used as material for analysis, for example for product design, marketing, product launch, service quality, level of competition, and others. In computer science, Natural Language Processing (NLP) & Machine Learning (ML) such as RNN can be used to predict implicit emotions. In this study, we apply a model to predict implicit emotions from audience comments using the RNN GRU algorithm. Comments that had been labelled emotional were then processed by improving words, replacing words with root words, and reducing words that did not provide a predictive performance improvement. Furthermore, with One-hot Encoding and Multilabelbinarizer, the data is converted into tensors which can then be processed by RNN GRU. The final results show that the model can work equally or better than other algorithms with the same dataset.
Keywords: Emotional Recognition, Natural Language Processing, Deep learning, Text Processing, Recurrent Neural Network
