MetricsAll MetricsMulti-Turn

Knowledge Retention

Knowledge Retention is a multi-turn metric to determine if your chatbot remembers details well.

Overview

The knowledge retention metric is a multi-turn metric that uses LLM-as-a-judge to evaluate whether your chatbot remembers important information provided by the user throughout the conversation.

Required Parameters

These are the parameters you must supply in your test case to run evaluations for knowledge retention metric:

turns
list of TurnRequired

A list of Turns as exchanges between user and assistant.

Parameters of Turn:

role
user | assistantRequired

The role of the person speaking, it’s either user or assistant

content
stringRequired

The content provided by the role for the turn

How Is It Calculated?

The knowledge retention metric first uses an LLM to extract information from the content of all user turns, then uses the same LLM to check if any assistant turns contain content that fails to recall this information.


Knowledge Retention=Number of Assistant Turns without Knowledge AttritionsTotal Number of Assistant Turns\text{Knowledge Retention} = \frac{\text{Number of Assistant Turns without Knowledge Attritions}}{\text{Total Number of Assistant Turns}}

The final score is the proportion of assisant turns with knowledge attrition found in the conversation.

Create Locally

You can create the KnowledgeRetentionMetric in deepeval as follows:

1from deepeval.metrics import KnowledgeRetentionMetric
2
3metric = KnowledgeRetentionMetric()

Here’s a list of parameters you can configure when creating a KnowledgeRetentionMetric:

threshold
numberDefaults to 0.5

A float to represent the minimum passing threshold.

model
string | ObjectDefaults to gpt-4.1

A string specifying which of OpenAI’s GPT models to use OR any custom LLM model of type DeepEvalBaseLLM.

include_reason
booleanDefaults to true

A boolean to enable the inclusion a reason for its evaluation score.

async_mode
booleanDefaults to true

A boolean to enable concurrent execution within the measure() method.

strict_mode
booleanDefaults to false

A boolean to enforce a binary metric score: 0 for perfection, 1 otherwise.

verbose_mode
booleanDefaults to false

A boolean to print the intermediate steps used to calculate the metric score.

This can be used for multi-turn E2E

Create Remotely

For users not using deepeval python, or want to run evals remotely on Confident AI, you can use the knowledge retention metric by adding it to a single-turn metric collection. This will allow you to use knowledge retention metric for:

  • Multi-turn E2E testing
  • Online and offline evals for traces and spans