Open AccessNEURAL NETWORK APPROACH FOR EMOTIONAL RECOGNITION IN TEXT
Author(s) -
Dmytro Nazarenko,
Iryna Afanasieva,
Nataliіa Golіan
Publication year - 2019
Publication title -
bionika intellekta
Language(s) - English
Resource type - Journals
eISSN - 2663-306X
pISSN - 2663-3051
DOI - 10.30837/bi.2019.1(92).02
Subject(s) - computer science , pooling , task (project management) , artificial neural network , dropout (neural networks) , artificial intelligence , backpropagation , component (thermodynamics) , metric (unit) , process (computing) , field (mathematics) , machine learning , embedding , recurrent neural network , natural language processing , engineering , operations management , physics , mathematics , systems engineering , pure mathematics , thermodynamics , operating system
The article is devoted to one of the most popular trends in the field of IT today – natural language processing, in particular, the extraction of emotions from the text using the neural network approach. The main task was to solve the problem of the high costs of time and human resources for companies to receive feedback from users and process emotional reactions of the second one. That to decide the task it was necessary to make modelling and learn neural network using own architecture based on the backpropagation algorithm that to recognize the emotional component in the text.The emotional component of reviews was used as a metric for evaluating user reactions. It was decided to work with five types of emotions that will help to provide better results. The neural network architecture consists of interconnected layers: embedding, bidirectional LSTM, pooling, dropout layers and two dense layers. For the neural network learning was selected an open dataset consisted of 47,288-tagged posts from Twitter. As a result, the F-measure on the test dataset was 0.62 and which is a worthy indicator in comparison with large business solutions.