Since I often find myself talking past customers and external engineers, I eventually realized that everyone has a different definition of chatbots. Did you know that Chatbots can be divided into three categories? In the first article of the Conversational AI column, we will introduce the ‘uses of various Chatbots’ and conduct an in-depth comparison regarding ‘development methods’, ‘features’, ‘key evaluation metrics’, and ‘application scenarios’, helping you fully understand Chatbots and avoid having only a partial understanding.
| Category | Question-Answering System | Task-Oriented Dialogue System | Chit-Chat System |
| English | Question Answering system | Task-Oriented Dialogue system | Chit-Chat Dialogue system |
| Function | Answer user questions | Complete tasks on behalf of users | Chat with users for companionship |
| Domain | Specific domain | Specific domain | Open domain |
| Method | Web search-based, knowledge base-based, community-based | Modular (rule-based, data-driven), end-to-end (data-driven) | Retrieval-based, generation-based |
| Features | Single-turn dialogue, focusing on question analysis (identifying information words) | Multi-turn dialogue, focusing onDialogue management | Multi-turn dialogue, focusing on personalization and sentiment analysis |
| Key Metrics | Recall, Precision, F-Measure | Task completion rate, dialogue duration, number of dialogue turns, machine-simulated user ratings | Word overlap rate, word vector distance, machine-simulated user ratings |
| Application Scenarios | FAQ, education | Assistant, ticket booking | Chit-chat, companionship |
| Notable Examples | IBM Watson | Siri、Google Assistant | Microsoft Xiaoice, SimSimi |
| Implementation Method | Knowledge base-based question-answering system | Modular task-oriented dialogue system | Applying search engine retrieval, or trainingSeq2seq modelGeneration |
