A Hybrid Semantic Analysis Approach Using Rule Based and Learning Techniques for Human-Robot Interaction in a Robotic Assistant

摘要

With the advancement of natural language processing techniques, robots are being built to interact in natural languages, in order to facilitate a more human like experience for its users. This study focuses on building a human-robot interaction system for a document delivering robotic assistant that is capable of following instructions and answering questions related to the domain of an office environment. The semantic analysis carried out in this closed domain intelligent system employs a hybrid approach comprised of both rule based and machine learning based techniques, while addressing the problem of limited domain related training data. Speech inputs obtained through speech recognition libraries are sent through several layers of classifiers in order to extract keywords that maps the input to the intended meaning of the user. Response templates backed by a knowledge base is used to generate the appropriate responses, which will then be converted into speech form. Apart from the rules utilized, machine learning based approaches SVM, KNN, Na?ve Bayes and Decision Trees are evaluated in this research for the semantic analysis classifications. The results of this hybrid approach show a significant level of accuracy and provides a solution to the shortage of domain related training data that limits the ability to use neural network based techniques.