CollegeBot: A Conversational AI Approach to Help Students Navigate College

Publication Date

10-17-2020

Document Type

Conference Proceeding

Publication Title

HCI International 2020 - Late Breaking Papers: Multimodality and Intelligence

Conference Location

Copenhagen, Denmark

DOI

10.1007/978-3-030-60117-1_4

Abstract

In an organization as big as a university that has many distinct departments and administrative bodies, it becomes almost impossible to easily obtain information online or by other means. Assistance over the phone or in-person is often limited to office hours and the information online is scattered through numerous (often nested) web pages, often independently administered and maintained by each sub-division. In this work, we present CollegeBot, a conversational AI agent that uses natural language processing and machine learning to assist visitors of a university’s web site in easily locating information related to their queries. We discuss how we create the knowledge base by collecting and appropriately preprocessing information that is used to train the conversational agent for answering domain-specific questions. We have evaluated two different algorithms for training the conversational model for the chatbot, namely a semantic similarity model and a deep learning one leveraging Sequence-to-Sequence learning model. The proposed system is able to capture the user’s intent and switch context appropriately. It also leverages the open source AIML chatbot ALICE to answer any generic (non domain-specific) questions. We present a proof-of-concept prototype for San Jose State University, to demonstrate how such an approach can be easily adopted by other academic institutions as well.

Keywords

Chatbot, Conversational AI, Natural language processing, Deep learning, Sequence-to-Sequence, AIML, Semantic sentence similarity.

Department

Computer Engineering

Link to Full Text

Share

COinS