Natural Language Understanding

The natural language understanding (NLU) component is responsible for obtaining a structured representation of text utterances. This typically entails dialogue act recognition; a dialogue act comprises an intent and its associated slot-value pairs. In addition to this, the NLU can also do satisfaction prediction. This is the user’s satisfaction with the agent’s response.

UserSimCRS is compatible with NLU components from the DialogueKit library, which provides basic NLU functionalities. Additionally, we implement usersimcrs.nlu.llm.LLMDialogueActExtractor, a dialogue act extractor based on a large language model.

LLM-based NLU

Prerequisites: the LLM used for the dialogue acts recognition should be hosted on a Ollama server.

The prompt used for the dialogue acts recognition should have the following placeholders:

• utterance: Placeholder for the text utterance to be processed.

Please note that the expected format for the dialogue acts recognition model output is a string representation of the dialogue acts, where each dialogue act is formatted as intent(slot1=value1, slot2=value2, …). Multiple dialogue acts are separated by a pipe (|). After parsing the model output, invalid dialogue acts are discarded.