Home // INTELLI 2017, The Sixth International Conference on Intelligent Systems and Applications // View article
Authors:
Gizem Soğancıoğlu
Tolga Çekiç
Bilge Köroğlu
Mert Basmacı
Onur Ağın
Keywords: chatbot dialog sentiment analysis
Abstract:
In this paper, we propose a goal-oriented chat bot, which aims to sell a suitable credit card to customers according to their needs. Our proposed chat bot detects customer's needs, uses this information to recommend a credit card, answers specific queries from customers and gathers the required information to start a credit card application process. This goal-oriented dialog management system is designed for Turkish Language and it makes use of a Finite State Machine (FSM) structure to achieve its goals. This design allows the chat bot to facilitate the flow of conversation and prevents giving unrelated answers to customers. The chat bot has unique tasks to perform in each state of FSM. Transitions between states are processed with the events, which are determined by outputs of the sentiment analysis model. Due to Turkish being an agglutinative language we perform morphological analysis of words to perform this task. Besides driving the conversation flow to achieving its goal, the chat bot can detect when customers ask questions and proceeds to the related state where the chat bot retrieves a proper answer. Since sentiment classification model forms the basis for keeping the chat bot in proper states, we experimented with different classification algorithms with different features and compared their successes. K-Nearest Neighbor algorithm using bag-of-words and lexicon features yielded the best results with the 0.822 f-score. Moreover, human evaluations for chat bot showed that using FSM and managing the conversation with the sentiment classification model for Turkish language is a promising solution.
Pages: 38 to 43
Copyright: Copyright (c) IARIA, 2017
Publication date: July 23, 2017
Published in: conference
ISSN: 2308-4065
ISBN: 978-1-61208-576-0
Location: Nice, France
Dates: from July 23, 2017 to July 27, 2017