Blooms Analyzer - Project for Pavatharini

Title: Automated Bloom’s Taxonomy Analyzer for University Question Paper Design using LLaMA 3.2

Abstract:

In outcome-based education (OBE), it is essential that university question papers are aligned with cognitive learning objectives to ensure holistic student development. Bloom’s Taxonomy, with its six hierarchical levels (K1–K6), provides a widely accepted framework for classifying educational objectives and assessing cognitive rigor. However, manual classification of exam questions into Bloom levels is subjective, time-consuming, and prone to inconsistencies. This research presents an AI-powered Bloom’s Taxonomy Analyzer that leverages the Ollama LLaMA 3.2 large language model to automatically classify university examination questions according to the cognitive levels defined in Bloom’s framework.

The proposed system uses prompt engineering techniques to guide the LLaMA 3.2 model in identifying the intent, action verbs, and cognitive complexity of each question. The analyzer accepts DOCX and PDF formats of question papers, extracts the content, and processes each question through a classification pipeline. It outputs Bloom level tagging (K1 to K6), identifies cognitive imbalances, and provides constructive feedback for question paper improvement. A curated dataset of university-level questions tagged by domain experts is used for validation, and performance is evaluated using standard NLP metrics such as precision, recall, and F1-score.

https://docs.google.com/document/d/12veZncuukii0N-_sFrJdhBjVcREZ10sN/edit?usp=drive_link&ouid=107023064021447656547&rtpof=true&sd=true

==========================================================================

Certainly! To create an agent that analyzes a question paper based on Bloom’s Revised Taxonomy (Knowledge Levels K1–K6) and provides remarks for improvement, we’ll break the process into two components:

✅ 1. Understanding Bloom's Revised Taxonomy (K1–K6)

Level	Cognitive Skill	Action Verbs
K1	Remembering	Define, List, Name, Recall, Identify
K2	Understanding	Explain, Summarize, Interpret, Describe
K3	Applying	Use, Implement, Solve, Demonstrate
K4	Analyzing	Compare, Organize, Deconstruct, Examine
K5	Evaluating	Judge, Critique, Defend, Justify
K6	Creating	Design, Construct, Develop, Compose

✅ 2. Agent Responsibilities

🎯 Core Functions of the Agent:

Input: A text-based question paper.
Process:
- Break questions into individual items.
- Identify the key action verbs and context.
- Classify each question into K1–K6 based on Bloom’s taxonomy.
- Check distribution balance across knowledge levels.
Output:
- Annotated questions with assigned Bloom’s level.
- Summary of question distribution across K1–K6.
- Critical remarks & suggestions for improvement.

✅ 3. Sample Workflow of Agent

✳️ Example Input:

1. Define thermodynamics.  
2. Explain the laws of thermodynamics with examples.  
3. Solve the given numerical based on the second law.  
4. Analyze the working of a Carnot engine.  
5. Justify the need for entropy in systems.  
6. Design an experiment to verify the third law.

✅ Output:

Q.No	Question	Bloom’s Level	Remarks
1	Define thermodynamics.	K1 – Remembering	Basic recall. Suitable for introductory section.
2	Explain the laws...	K2 – Understanding	Good for conceptual clarity.
3	Solve...numerical	K3 – Applying	Promotes procedural knowledge.
4	Analyze...Carnot engine	K4 – Analyzing	Encourages critical breakdown.
5	Justify...entropy	K5 – Evaluating	Promotes higher-order thinking.
6	Design...experiment	K6 – Creating	Highest cognitive level. Excellent.

📊 Bloom’s Level Distribution:

Level	Count
K1	1
K2	1
K3	1
K4	1
K5	1
K6	1

💬 Remarks:

✅ Excellent distribution across all levels.
⚠️ Ensure balance with course outcomes (COs).
✅ Higher-order thinking (K4–K6) well represented.
🔁 May increase weightage on K3 if application skills are a program outcome.

✅ 4. Implementation Plan for Agent (Tech Stack)

🛠️ Tools & Technologies:

Language Model: ChatGPT / GPT-4.5 or OpenAI with Bloom’s logic embedded
Frontend (optional): Web UI (React or Angular)
Backend Agent Workflow: Python (with LangChain or custom NLP)
File Input: PDF / DOCX / TXT parser (e.g., pdfplumber, python-docx)

📄 Pseudocode Overview:

for question in question_paper:
    extract_verb = identify_action_verb(question)
    level = map_to_blooms_taxonomy(extract_verb)
    assign_level(question, level)

generate_distribution_chart()
analyze_distribution()
generate_remarks()

✅ 5. Would You Like a Working Code Example or UI Mockup?

Let me know if:

You want a Python script that reads a question paper and returns a table.
You want a UI-based tool (React or Angular).
You want integration with curriculum/course outcome mapping.

Would you like to upload a sample question paper for analysis? I can demonstrate the full pipeline live here.

=========================================================================