Conversational Interfaces is an interactive agent that conducts conversation via textual and auditory mode. The CI are of two types and they are Chatbots & Virtual Assistants (Eg: Siri, Google Now). Chatbots are at the ultimate of developing area. It is one of the most useful technologies that replacing the traditional models and making apps and websites inessential. The most interesting feature of the bots is that they learn from the previous interactions and become smarter. Conversational models works in two ways- rule based and smart machine based. Rule based models follow rules to do job and smart machine based models use machine learning to do job and adapt their behavior based on experience.
Chatbot are the software program that conduct conversation via textual and auditory mode. The technology used to do the job by the Chatbot can be grouped into two categories. First one is the rule based work that use rules and heuristics to do its job. Alternate one is the smart machine based work, where they use machine learning to do work and can learn on their own.
While planning a framework for Chatbot, we have to consider mainly two things – the questions should come either under open domain or closed domain and the responses from the Chatbot should come either under retrieval based system or generative based system. In closed domain there will be a limited to as a set of questions on specific topics. The open domain deals with questions that can come under any topics. Therefore it is clear that open domain is very difficult when compared to closed domain because the customer can take the conversation anywhere. The Chatbot needs to generate responses once a question asked. In retrieval based system, the Chatbot contains a set of predefined responses and the generative based system doesn’t have predefined response but rather generate responses from scratch and the knowledge already acquired.
Today mostly all the generative conversational agent are build using deep learning. By using the deep learning, it is easy to generate responses from the past interactions and current input. Thus generative model is a conversational agent which generate responses from the previous conversations between the humans.
Generative models are better than retrieval – based models. They can generate the responses and not always replies with one of the answers from a set of answers. This makes them more intelligent as they take word by word from the query and generates the responses. It also makes them more prone to errors as they need to take the spelling and grammar into account. Once trained, they outperform the retrieval – based models as they can answer complex and unseen queries using Language translation models. Once a question asked from closed domain, the Chatbot generate responses using smart machine technology. Thus the Chatbot can handle common questions and the unforeseen cases without predefined responses.