Chatbots are designed to interact with users. Users submit messages (text/voice) and depending upon the user’s “intent” behind the message, the chatbot formulates a relevant and tailored response. The “intent” is the user’s reason behind engaging with the chatbot. One complexity is that users may share an identical “intent” but express that intent differently when submitting messages to a chatbot. By understanding the “intent” behind messages, chatbots need not rely on specific and consistent user inputs.
Chatbot intent training
Chatbots can be trained to recognize various types of user intents including Navigational, Informational, Transactional, Feedback and other types that may be unique to an enterprise.
The purpose of the chatbot is defined by the enterprise. To align with enterprise goals, chatbot “intent training” requires planning, the thoughtful development of training data, multiple iterations to fine turn the responses generated by chatbots, and continuous improvement to deliver a satisfying user experience.
Chatbot intent training includes:
- Define the chatbot’s purpose: Identify the goal(s) of the chatbot as it aligns with the enterprise and determine the types of problems that the chatbot will address.
- Conduct user research and analysis: Determine the types of questions or messages that the chatbot may receive from users. Consider variations of phrases or text for these questions or messages that users may use.
- Create the intents: Create and group intents that align with the chatbot purpose.
Note: Invest time identifying different ways users may pose queries so that chatbots can match a greater variety of queries to a specific intent.
- Fabricate the training data (annotation): Annotate training data (e.g., chatbot text conversations). This identifies key components including intent and entities that are encapsulated within the training data text.
- Train the chatbot: Use the annotated data (step 4) to train the chatbot. The model will factor in the intents and entities identified in the training data to learn patterns and associations between inputs and corresponding intents. This will allow the chatbot to predict user intents when presented with new chatbot text conversations.
- Test and monitor: Test the chatbot by introducing new chatbot text Determine if the chatbot responses match real user queries and fine-tune the chatbot responses as required to ensure that the chatbot responses are efficiently fulfilling user needs.
- Update training data: The training data may become dated, or it may require refinement through the addition of new intents to deliver better user experiences. This is an iterative process improvement that should receive regularly scheduled reviews.
