Intelligent algorithms for social robotics

Abstract

This project aims to evaluate the conversational abilities of a social robot, Nadine. The analysis was done by analysing the text data collected from her conversations with humans, when Nadine was “working” as a receptionist at ArtScience Museum. As sentiment is an element that highly affects the quality of the interaction between human and social robots, this project focuses on sentiment analysis, through two approaches. The sentiment polarity of Nadine’s and human’s speeches were classified using Naïve Bayes (NB) classifier. This part of analysis presents the percentage of negativity in the speeches of both parties. Nadine’s sentiment in the conversations was determined to be slightly more negatively inclined than human’s. As a numerical measure is needed to evaluate the degree of positivity and negativity in their speeches, the sentiment score of each speech spoken by Nadine and human was obtained using the SentiWordNet lexicon-based approach. From there, speeches with the most positive or negative sentiment were obtained. To improve the quality of the interaction between Nadine and human, the sentences with highly negative sentiment scores are suggested to be avoided or rephrased into more positively inclined sentences. The findings drawn from our analysis serves as a reference for improving the conversational systems of Nadine or other social robots.