Sentiment classification of delta robot trajectory control using word embedding and convolutional neural network

Abstract

Sentiment classification (SC) is an important research field in natural language processing (NLP) that classifying, extracting and recognizing subjective information from unstructured text, including opinions, evaluations, emotions, and attitudes. Human-robot interaction (HRI) also involves natural language processing, knowledge representation, and reasoning by utilizing deep learning, cognitive science, and robotics. However, sentiment classification for HRI is rarely implemented, especially to navigate a robot using the Indonesian Language which semantically dynamics when written in text. This paper proposes a sentiment classification of Bahasa Indonesia that supports the delta robot to move in particular trajectory directions. Navigation commands of the delta robot were vectorized using a word embedding method containing two-dimensional matrices to propose the classifier pattern such as convolutional neural network (CNN). The result compared the particular architecture of CNN, GloVe-CNN, and Word2Vec-CNN. As a classifier method, CNN models trained, validated, and tested with higher accuracy are 98.97% and executed in less than a minute. The classifier produces four navigation labels: right means 'kanan', left means 'kiri', top means 'atas', bottom means 'bawah', and multiplier factor. The classifier result is utilized to transform any navigation commands into direction along with end-effector coordinates.