Comprehensive Guide: Creating a Q&A Dataset Using the Data Source Module

For QA Testing, Content Validation, and Bot Answer Evaluation

1. Overview

The Data Source module allows you to upload structured spreadsheet files (CSV or Excel) and automatically generate Question–Answer (Q&A) pairs.
This capability is designed for:

Bot Training & Evaluation – Quickly testing whether your conversational bot answers questions accurately.
Content Validation – Verifying that Q&As generated from tabular or structured content are correct.
Knowledge Base Building – Creating structured Q&A content without manually writing every question.

This document explains, in detail, how QA testers and content managers can upload CSV or Excel datasets, review generated questions and answers, and finalize them for testing.

2. Key Concepts & Terminology

Term	Explanation
Dataset	A collection of Q&A pairs generated from one uploaded spreadsheet.
Q&A Pair	A single question with its corresponding answer extracted from the spreadsheet.
CSV/Excel File	A spreadsheet containing the source data from which the system will generate Q&As.
Status: Approved / Rejected	Each generated Q&A can be marked Approved (usable for training/testing) or Rejected (not suitable).
Bot Space	The workspace in the application where you manage your bot’s data sources, training sets, and testing tools.

3. Prerequisites

Before you begin, ensure:

You have login access to the application and permission to view and use the Bot Space and Data Source module.
Your data is stored in a CSV (.csv) or Excel (.xlsx/.xls) file. Note: PDFs or image-based documents are not supported for dataset creation.
The spreadsheet should contain clearly defined columns and rows that can be used to form Q&A pairs (for example, “Question” and “Answer” columns, or structured tabular content).

Tip: Clean, well-labeled columns produce more accurate Q&A results.

4. Dashboard Metrics

At the top of the Data Source page, the system displays key indicators:

Total Datasets – Total number of datasets you’ve created.
Datasets Processing – Number of datasets currently being analyzed.
Datasets Completed – Datasets successfully processed and ready for review.
Total Q&A Generated – Total number of Q&A pairs generated across all datasets.

These metrics help you monitor system usage, spot processing delays, and manage testing capacity.

5. Step-by-Step Process

Step 1: Access the Data Source Module

Log in to the application.
From the left navigation menu, click Bot Space.
- This is the central hub for all bot-related configurations and testing.
Inside Bot Space, click Data Source.
Review the dashboard metrics to confirm system status.

Below the metrics, you’ll see a Dataset Table with columns:

No. – Serial number.
Dataset Name – Custom name given by you (e.g., “Customer Support FAQs”).
Question Type – “Document” (indicates a file-based source).
# of Q – Total questions generated from the uploaded spreadsheet.
Date & Time – When the dataset was processed.
Actions – Icons to View, Edit, Delete, or Download.

Step 2: Create a New Dataset

Click + Create New (next to the search bar).
In the Create Dataset popup:
- Name: Enter a meaningful name (e.g., “Product FAQs – October”).
- Add Document: Click to browse or drag-and-drop your CSV or Excel file.
  - Supported formats: .csv, .xls, .xlsx.
  - Ensure the file has clear headers (for example, Question, Answer, Category).
Click Process (green button).
- The system uploads the file, reads each row, and generates Q&A pairs.

Behind the scenes:
Each spreadsheet row is parsed; depending on column headers, the system determines how to form each question and answer.

Step 3: Review and Approve Q&A Pairs

Once processing is complete:

Your new dataset appears in the main table with its generated # of Q.
Click the View (eye) icon to open details.

You will see:

No. – Sequential question number.
Question – System-generated question text (from the “Question” column or auto-derived).
Answer – Corresponding answer from the spreadsheet.
Status – “Approved” or “Rejected.”

Actions within the detail view:

Filter By: Narrow the list by status (Approved / Rejected).
Search Answer: Locate specific text quickly.

QA Tip:

Approve only rows where the generated Q&A exactly matches the intent of the spreadsheet content.

Reject duplicates or incomplete Q&As.

Step 4: Edit Dataset (Optional)

Back in the main table, click the Edit (pencil) icon.
Update the dataset Name or upload a revised spreadsheet if changes are required.
Click Process again to regenerate Q&As.

Step 5: Additional Actions

Delete (trash icon): Permanently remove the dataset.
Download (down-arrow icon): Export the approved Q&A list as a CSV or Excel file for offline review or integration.
Submit / Reset: In the dataset detail view, use Submit to save all status changes or Reset to clear applied filters.

6. Best Practices for QA Testing

Spreadsheet Structure
- Use clear headers (e.g., Question, Answer).
- Avoid merged cells or inconsistent formatting.
Data Cleanliness
- Remove duplicate questions before uploading.
- Ensure consistent spelling and grammar for better NLP processing.
Validation
- Check that every approved Q&A precisely reflects the spreadsheet data.
- Test the bot by asking a random selection of approved questions.
Performance Monitoring
- Large spreadsheets may take longer to process.
- Use the dashboard metrics to track processing times.
Error Handling
- If the system flags a file format issue, confirm the file is indeed CSV or Excel and properly encoded (UTF-8 is recommended).

7. Example QA Workflow

Prepare an Excel file with columns Question and Answer.
- Example row:
  - Question: Who is the chairperson of Mumbai Port Authority?
  - Answer: Dr. Rajiv Jalota.
Upload the Excel file in the Data Source module.
Review the automatically generated Q&As.
Approve accurate entries and reject any that need correction.
Download the approved dataset and use it to test the bot by asking those questions.

8. Conclusion

The Data Source module provides a structured way to create Q&A datasets exclusively from CSV or Excel spreadsheets.
By following this guide, QA teams can:

Validate content accuracy against the original spreadsheet.
Evaluate bot performance with real, structured data.
Maintain a high-quality, up-to-date knowledge base for conversational AI.

Document Owner: Arjun Haldankar
Version: 1.0
Last Updated: 25-09-2025

Product KT Document

Data Source