A General Framework for Constructing Domain-specific Chatbots
- Ttitle : A General Framework for Constructing Domain-specific Chatbots
- Author : Wenjing Yang, Cheng Zhang, Jie Wang
- Affiliation : Department of Computer Science, University of Massachusetts Lowell
- Journal : International Journal of Applied Informatics and Media Design
- Volume : Vol.1
- ISSN : 2436-8814
- ISBN : 978-4-908181-47-4
- Pages : pp.7-12
- Keywords : Chatbot, Information Retrieval, Question-Answer System, Text Classification, Word Mover’s Distance
Abstract : This paper presents an easy-to-deploy general framework for chatbot builders to construct domain-specific chatbots called dChat, which performs the following tasks: (1) Collect domain-specific query-answer pairs (QAPs) from community-based question-answering websites and generate new QAPs suitable for human-chatbot conversations. (2) Con- struct a chatbot for a specific domain. (3) Switch domains based on user queries. In particular, dChat improves previous work by adding domain classifications and Word2Vec keyword searches. Comparing SVM and CNN classifiers trained on domain-specific QAPs we collected, we show that SVM provides much higher F1 scores on user queries. We then apply WMD (Word Mover’s Distance) to find semantic relationships between queries and search for the most appropriate answer to the user query. Experimental results show that dChat achieves an 8.4% increase of accuracy over previous results on selecting answers.