Skip to content

sobala/codes_performance_analysis

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 

Repository files navigation

AI Model Confirmation Analytics

This project analyzes how users interact with and confirm outputs from an AI model.
It focuses on understanding both model performance and user trust behavior through multiple confirmation metrics.


📊 Project Overview

The system allows users to attempt up to three different questions, each with multiple attempts (new version of a platform is restricted to only 3 attempts).
Each attempt can result in a confirmation (True/False) by the user depending on whether the user agrees with the classification response received by the AI model. The project evaluates:

  • How often users confirm the model’s answers
  • How user behavior differs from model correctness
  • Why users sometimes fail to confirm or provide ingenuine inputs

🎯 Objectives

  1. Measure user confirmation rates at different levels:

    • Attempt-level: % of individual attempts that were confirmed.
    • Question-level: % of user–question pairs that were ever confirmed.
    • User-level: % of users who confirmed at least one question.
  2. Evaluate model accuracy, independent of user confirmation:

    • Whether the model’s predicted category matches the correct category.
    • Comparison between CategoryCorrect (objective correctness) and IsUserConfirmed (user perception).
  3. Analyze non-confirmation reasons, such as:

    • Model error (model_wrong)
    • User confusion (user_confused)
    • Ingenuine input (user_ingenuine)
    • Old labels or outdated categories (OldLabel)

🧮 Data Structure

Each row in the dataset represents one attempt by a user.

Column Description
SurveySessionID Unique session identifier corresponding to a unique user
QuestionID Unique question/task identifier
AttemptID Sequential attempt number for that question
IsUserConfirmed Whether the user confirmed the response (True/False)
CategoryCorrect Whether the model prediction was objectively correct
UserInputCorrect Whether the user input was genuine or valid
FailureReason Reason for failure if not confirmed
SubclassPresent Whether the subclassification was provided

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors