Combining TF-IDF and LDA to generate flexible communication for recommendation services by a humanoid robot

Abstract

Linguistic flexibility around non-predetermined expressions is important for more effective human–robot face-to-face interaction. In the past, most robots have been fitted with a limited and supervised response process and programmed with certain responses for predetermined words or sentences. As a result, implementing viable robot-based recommendation services has been difficult. The purpose of this paper is to propose a text-mining approach to flexible robot-based recommendation services in which, when the robot encounters linguistic expressions that differ substantially from the programmed linguistic database, flexible responses are generated based on understanding of several external corpora and a knowledge-learning process. This study combines two text-mining methods, TF-IDF and LDA, to generate flexible communication, which enables a robot to respond with recommendation content that is not pre-programmed. The results of our analysis suggest that the proposed combined approach outperforms the TF-IDF and LDA methods in terms of overall accuracy and F-score.