Dialog management systems (DMS) are the natural language processing (NLP) components responsible for interpreting and contextualizing human-like conversations between chatbots and live users. For example, a user might ask to place a coffee order, and the chatbot would respond by taking the order. However, the user later in the interaction might say, “change it to a bagel,” where the chatbot then becomes responsible for interpreting the request and adjusting the order correctly before confirming the customer’s final order. Dialog management systems, or plug-ins, enable the chatbots to perform this task without issue.
The dialog management system has a module that enables the DMS to contextualize information and provide responses called the agent for dialog management. The agent has two distinct functions: policy learning and a feedback mechanism from the environment.
Policy learning is a high-level framework that teaches the bot positive language, similar to a customer service representative, which increases user satisfaction. The bot uses the policy learning framework to learn from each interaction while referring to similar past conversations in which it has participated.
The feedback mechanism exists to collect feedback on how the bot is performing from the end-users. Critical feedback reinforces the bots learning process, forcing it to learn from the mistakes it has made.
Very few NLP systems use a collective response and feedback model for appropriate response generation. User-friendly bots, however, use historical response data to make response adjustments that are more likely to increase user satisfaction. Often, dialog managers need to make a trade-off between instant and long-term satisfaction by collecting more response data before adjusting.
Dialog manager control flow for a chatbot.
Response models are programmed to use various tactics to produce output responses on numerous topics. However, the dialog manager is first responsible for generating an appropriate answer for the conversation at hand. The dialog manager uses historical chat data and compiles various possible outputs using different response models.
Chatbots also have an internalized natural language understanding (NLU) unit, which interacts with the DMS, generating the dialogue task’s semantic representation. At this point, the DMS becomes responsible for all regulations regarding the conversational structure between the chatbot and end-user. The DMS regularly receives input from the NLU components to maintain an interface with the task managers before passing the output to the natural language generator (NLG).
