HUMAN AND ROBOT DIALOGUE MANAGEMENT SYSTEMS USING TF IDF COSINE SIMILARITY AND JACCARD COEFFICIENT

• Main Author: Ayu Permatasari, Dinda
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/35576
• Institution: Institut Teknologi Bandung

Dialogue management is a system of software and hardware designed to decide the actions that must be carried out by robot. This dialog management system allows users to be able to interact with natural languages using speech. TF-IDF method is used to give weight to the relationship of word terms, while the Cosine Similarities and Jaccard Coefficient are used as a comparison method to determine the classification of user sentence intentions based on sentence similarity in management dialogue. Finite State Machine method to managing the flow along with the sequence of dialogs that have been determined by scenario to determine the actions that must be performed by the robot. In Indonesian speech recognition process, Google Cloud Speech API is used as an engine to convert speech into text using Kinect 2.0 as a sound sensor. Results of the research that have been carried out show that the level of accuracy for the use of speech recognition shows 97.25% with a translation time response changing speech into text with a range between 2.32 s - 3.0275 s. Classification of sentences using the Cosine Similarity in accordance with the dialogue scenario shows an average accuracy of 99.68%, and sentences with different words from the dialog scenario show an average accuracy of 98.75%. Classification of sentences according to the scenario of dialogue using the Jaccard coefficient shows an accuracy of 98.44%, whereas the question sentence with words that are different from the dialog scenario shows the accuracy reached 94.29%. In this study, TF-IDF Cosine Similarity method has a better performance than the Jaccard coefficient.