The ConversationCompletenessMetric is a conversational metric designed to evaluate whether an LLM chatbot can complete an end-to-end conversation by satisfying user needs throughout the interaction. It serves as a proxy to measure user satisfaction in conversational AI applications.

​
Required Arguments

  • turns: A list of ModelTestCase instances representing the conversation turns. Each ModelTestCase requires:
  • input: The user’s input to the chatbot.
  • actual_output: The chatbot’s response to the user’s input.

​
Optional Arguments

  • threshold: A float representing the minimum passing threshold, defaulted to 0.5.
  • model: A string specifying which of OpenAI’s GPT models to use, or any custom LLM model of type DeepEvalBaseLLM. Defaulted to ‘gpt-4o’.
  • include_reason: A boolean which, when set to True, includes a reason for its evaluation score. Defaulted to True.
  • strict_mode: A boolean which, when set to True, enforces a binary metric score: 1 for perfection, 0 otherwise. It also overrides the current threshold and sets it to 1. Defaulted to False.
  • async_mode: A boolean which, when set to True, enables concurrent execution within the measure() method. Defaulted to True.
  • verbose_mode: A boolean which, when set to True, prints the intermediate steps used to calculate the metric to the console. Defaulted to False.

​
Usage Example

from agensight.eval.test_case import ModelTestCase, ConversationalTestCase
from agensight.eval.metrics import ConversationCompletenessMetric

convo_test_case = ConversationalTestCase(
    turns=[
        ModelTestCase(input="Hello, how can I help you today?", actual_output="I need assistance with my account."),
        ModelTestCase(input="Sure, I can help with that. Can you please provide your account number?", actual_output="My account number is 123456.")
    ]
)
metric = ConversationCompletenessMetric(threshold=0.5)

metric.measure(convo_test_case)
print(metric.score, metric.reason)