Abstract
Intelligent conversational computer systems, known as chatbots, have always been at the forefront of artificial intelligence. They are made to sound like humans in order for machines to communicate with humans. Because of the rising benefits of chatbots, numerous sectors have adopted them to give virtual support to clients. They are also used as companions and virtual assistants. Natural language processing and deep learning are two artificial intelligence disciplines that are used in chatbots. This chapter will examine current advancements in chatbot algorithms, approaches, and technologies that use artificial intelligence and/or natural language processing.
TopBackground
Despite the popularity of chatbots, creating chatbots that deliver satisfactory responses to the requirements of specific users remains an arduous task. For example, a chatbot must understand any user's speech or text as an input request and respond appropriately (e.g., on the same topic, make sense), helpfully (e.g., contains useful and concrete information), and even be tone-aware (e.g., conveys feelings like empathy and passion) (Xu et al., 2017; Hu et al., 2018).
A common way to design chatbots is the rule-based method (Young et al., 2013; Mesnil et al., 2015). It defines the structure of a dialogue state as a series of slots to be filled throughout a discussion. The chatbot responses depend on certain hand-crafted rules.
However, one of the main limitations of rule-based chatbots is that they are domain dependent. Each set of rules developed applies only to a limited number of cases and a limited field.
The alternative approach to rule-based chatbot design is data-driven techniques, also known as Machine Learning techniques. This design approach may handle more diverse user inquiries than rule-based chatbots (Song et al., 2018). The main types of data-driven chatbots are Information retrieval-based and Generative chatbots. Information retrieval-based chatbots may retrieve an existing response to users' queries from a pre-compiled dataset. Generative chatbots build a new response word by word depending on the input sequence provided by the user (Yan et al., 2016; Serban et al., 2016). One of the main strengths of information retrieval-based chatbots is they can provide the user with highly accurate responses. But they are restricted by the size of the corpus because they cannot develop new responses. Generation-based chatbots may be able to solve this problem. But they are prone to providing grammatically incorrect or useless replies (Song et al., 2018).
This chapter will focus on providing an overview of chatbot implementation methods. This will include recent technologies, algorithms, techniques, and evaluation methods. A distinction will be drawn between two approaches to chatbot design: Rule-based chatbots and Machine Learning (ML) based chatbots. Within ML-based chatbots, a further distinction will be drawn between Information-Retrieval chatbots and Generative Chatbots. A distinct section will be dedicated to transformers and transformer-based chatbots, as these are the most recent algorithms applied to the problem of Dialogue Modelling. The last section of this chapter will explore evaluation techniques and differentiate between automatic evaluation metrics and human evaluation metrics and techniques.
Key Terms in this Chapter
Recurrent Neural Network (RNN): A subset of artificial neural networks where the connections between the nodes can cycle, allowing the output from one node to influence the input to another node. It can display temporal dynamic behavior because of this. RNNs, which are derived from feedforward neural networks, may process input sequences of different lengths by using their internal state (memory).
Natural Language Processing (NLP): The analysis and synthesis of natural language and speech using computational methods. Building machines that comprehend and react to text or voice data—and answer with text or speech of their own—much like humans do is the goal of natural language processing.
Information Retrieval (IR): The process of locating information system resources that are pertinent to a particular information demand from a collection of those resources in computing and information science. Searches may use full-text indexing or another type of content-based indexing.
Dialogue Modeling: An abstract representation of how a user interacts with an interactive computer system. The notations used in user interface management systems are based on dialogue models (UIMS).
ChatBot: A computer program or software application created to mimic communication with human users, particularly online, through text or voice.
Transformer: A deep learning model that uses the self-attention process and weights the importance of each component of the input data differently.
Human-Computer Interaction (HCI): An interdisciplinary branch of research devoted to the design of computer technology and the interaction between people (the users) and computers.