Ticketing Chatbot Service using Serverless NLP Technology Eko Handoyo Department of Electrical Engineering Diponegoro University Semarang, Indonesia eko_handoyo@elektro.undip.ac.id Maman Somantri Department of Electrical Engineering Diponegoro University Semarang, Indonesia mmsomantri@live.undip.ac.id M.Arfan Department of Electrical Engineering Diponegoro University Semarang, Indonesia arfan@elektro.undip.ac.id Aghus Sofwan Department of Electrical Engineering Diponegoro University Semarang, Indonesia asofwan@elektro.undip.ac.id Yosua Alvin Adi Soetrisno Department of Electrical Engineering Diponegoro University Semarang, Indonesia yosua@live.undip.ac.id Enda Wista Sinuraya Department of Electrical Engineering Diponegoro University Semarang, Indonesia enda_sinuraya@elektro.undip.ac.id Abstract— Personal assistant using a human operator need some time to process single request such as ticket booking, ordering something, and get services. One request can contain many queries for some information provided on the internet. Business performance values time efficiency so must be considered an alternative way to take request. Chatbot can give 24 hours service which can become an advantage besides using a human personal assistant. Chatbot acts like routing agent that can classify user context in conversation. Chatbot helped with natural language processing (NLP) to analyze the request and extract some keyword information. One important process in NLP is morphological analysis and part of speech (POS) tagging. POS help to parse the meaning of chat text based on a set of rules. The rule base is specific to some language and designed to capture all the keyword relies on chat text. Keyword in booking conversation term is like departure and destination city and also the date of flight. There is a variation from a user determining city and date. NLP in booking confirmation has a task to analyze various pattern describing ordering requests like city and date. Messenger bot would be an example of assistance that can help user connected to many services some like ticketing service through conversation interaction. The contribution of this research is to conduct some scenario that happening in ordering tickets. This research conduct that chatbot can help acts as customer service, based on the conducted scenario and show an F- measure score of 89.65%. Keywords—chatbot, routing agent, conversation, NLP, interaction, intent I. INTRODUCTION Chat or speech is one meaningful form of communication between humans[1]. Chat becomes more natural interaction than graphic base interface so will be broadly used in humanizing computer interaction to human. Chatbot worked by interpreting the message that given by the user, and then give response base on captured parsed meaning of the message [2]. In 2015, Facebook as a social media platform allows some developer to build chat automation platform. This policy followed by another social chat platform like LINE and Telegram. Facebook provides a button and chats dialogue to help chat interaction especially to promoting some feature in chatbot. Chat dialogue also can be used to promoting product service and how to access those services. Chatbot interaction is the most important feature to design. Chatbot interaction must meet user need to the product. Although graphical based interface designed to be executed well, there is still some potential using chatbot. Icon clicking interaction sometimes misused because the user is not computers friendly enough [3]. Chat interaction can also be helped assisted routing to specific command if the user wrongly sent the request. NLP as the core of chat interaction is known build in cloud-based cognitive service. There are many services provided AI building blocks such as IBM Watson, Wit.AI, and Dialog Flow. Wit.AI is one of natural language interface for an application that capable turning sentence into structured data. The developer can integrate the service to well-known social media platform such as Facebook with the token. Wit.AI provide a built-in building block that can detect special intent word in a sentence. The developer has to handle coordination between cognitive service such as chatbot interface, integrate chatbot with third-party services, and also considering extensibility, scalability, and maintenance [4]. Serverless becomes a solution that let developer build function into the shared platform. Serverless build in a standard language with stateless technology. Stateless technology doesn’t store session information. Serverless programming model, deploy function that can be reach or executed from the cloud. Functions are stateless because of independence from previous runs. The function can invoke directly or triggered by some events. Chatbot application will utilize some function that can be arranged in conversational context [5]. In a conversational context, function chained together by sequence. In this work, Serverless model that used is Webhook. Webhook conducted to receive a direct message from the Facebook page. Facebook page connected to Wit.AI NLP service to get NLP features. Facebook send message response with NLP features such as location, intent, or number. NLP features parse to serverless function so can return some specific response. For specific location intent response, the Serverless function does a querying out a mechanism to external ticketing API to get price and information. The remainder of this paper is organized as follows. In Section 2 we provide methods of NLP and serverless programming model and how to integrate it to social media. In Section 3 we provide testing of several chatting scenarios. Proc. of 2018 5th Int. Conf. on Information Tech., Computer, and Electrical Engineering (ICITACEE) 978-1-5386-5529-0/18/$31.00 ©2018 IEEE 325