# Chatbot System Architecture

**Authors:** Moataz Mohammed, Mostafa M. Aref  
**arXiv:** [2201.06348](https://arxiv.org/abs/2201.06348)  
**Date:** 2022-01-17  
**Summary page:** https://www.summarizepaper.com/en/paper/2201.06348v1/

## Key points

- Computer science has seen a significant interest in conversational agents or chatbots
- Chatbots utilize Natural Language Processing (NLP) and Artificial Intelligence (AI) techniques
- The paper focuses on discussing the system architecture for conversational agents
- Various architectures for conversational agents were explored, including Amazon Alexa and Google Assistant
- The most commonly used generic architecture consists of three main components: Natural Language Understanding (NLU), Dialog Manager (DM), and Natural Language Generation (NLG)
- NLU includes subcomponents such as Topic Detection, Intent Analysis, and Entity Linking
- DM manages the conversation flow with subcomponents like Rule-based systems, Knowledge-based systems, Neural Network Reply Generation, and Online Information Retrieval
- NLG generates appropriate responses based on context with subcomponents like Content Filter and Engagement Ranking
- Evaluation was done through a survey at the end of each Messenger conversation to gather user feedback
- DM is a critical part of any chatbot system that can be continuously improved or alternative subs can be developed

## Summary

The field of computer science has seen a significant interest in conversational agents, also known as chatbots, over the past decade. These agents utilize Natural Language Processing (NLP) concepts and Artificial Intelligence (AI) techniques, such as Deep Learning methods, to generate appropriate responses. This paper focuses on discussing the system architecture for conversational agents and providing detailed explanations of each component. In the literature review, various architectures for conversational agents were explored. Examples include Amazon Alexa bot which utilizes Automatic Speech Recognition (ASR) and Amazon Web Services (AWS) to process user data and store conversations; and Google Assistant which breaks down voice recordings into individual sounds and matches them with similar word pronunciations stored in its database. Despite these diverse architectures, this paper focuses on the most commonly used generic architecture consisting of three main components: Natural Language Understanding (NLU), Dialog Manager (DM), and Natural Language Generation (NLG). The first component NLU plays a crucial role in understanding user input. It includes subcomponents such as Topic Detection, Intent Analysis, and Entity Linking. The second component is DM which manages the conversation flow. Its subcomponents include Rule-based systems, Knowledge-based systems, Neural Network Reply Generation and Online Information Retrieval. Finally NLG or Reply Generator generates appropriate responses based on context and may include subcomponents like Content Filter and Engagement Ranking. A simulation of these components is depicted in Figure 1 of the paper's source material. Moving on to evaluation a survey was conducted at the end of each Messenger conversation to gather feedback from users about their experience with the chatbot. Questions included inquiries about comfortability with the Messenger bot and whether users would recommend it to others. Responses were scaled from 1 to 10. The conclusion emphasizes that DM is a critical part of any chatbot system since its components can be continuously improved or alternative subs can be developed. The paper serves as a valuable reference for those interested in developing chatbot systems or enhancing the Dialog Manager components.