# CASML - Generative AI Hackathon — Data

来源：https://www.kaggle.com/competitions/casml-generative-ai-hackathon/data

采集时间：2026-10-19T13:16:28Z

# Kaggle RAG Hackathon: Dataset and Submission Format

## Overview

Welcome to the Retrieval-Augmented Generation (RAG) Hackathon! This competition challenges participants to develop advanced RAG systems capable of answering questions using a comprehensive psychology textbook as the knowledge base.

## Competition Objectives

- Develop a RAG system that can efficiently retrieve relevant context from an 800-page textbook
- Generate accurate and concise answers to a series of psychology-related queries
- Demonstrate advanced information retrieval and answer generation techniques

## Dataset Components

### 1. Queries File (`queries.json`)

The `queries.json` file contains the questions that participants must answer. Each query includes:

- `query_id`: A unique identifier for the question
- `question`: The specific query to be answered

#### Example Queries
- "What is the scientific method in psychology?"
- "What are the basic parts of a neuron?"
- "What are the stages of sleep?"

### 2. Submission File Requirements (`submission.csv`)

Participants must submit a CSV file with the following columns:

| Column     | Description                                                    | Type         |
|------------|----------------------------------------------------------------|--------------|
| ID         | Unique identifier matching the query_id                        | Integer      |
| context    | Retrieved text from the textbook used to answer the question   | String       |
| answer     | Generated response to the query                                | String       |
| references | JSON string indicating textbook sections and pages             | JSON String  |

## Submission Guidelines

### Key Requirements
- All queries must be answered
- Match query IDs exactly
- Provide context and references for each answer
- Use valid JSON formatting for references
- Include sections and page numbers for verification

### Reference Format Example
```json
{
  "sections": ["psychological_research/approaches_to_research"],
  "pages": [41, 3]
}
```

## Evaluation Criteria

Submissions will be assessed based on:
- Relevance and accuracy of retrieved context
- Quality and precision of generated answers
- Completeness of references
- Adherence to submission format guidelines

## Competition Tips

1. Optimize context retrieval mechanisms
2. Ensure answer quality and factual accuracy
3. Implement robust reference tracking
4. Test system thoroughly with sample queries
5. Focus on efficiency and scalability

## Resources

- Sample input: [Sample queries.json in Documentation]
- Sample output: [Sample submission.csv in Documentation]

## Final Submission Checklist

- [ ] All queries addressed
- [ ] Correct column names
- [ ] Valid data types
- [ ] No missing values
- [ ] Proper CSV and JSON formatting
- [ ] File named `submission.csv`

## Additional Information

For questions or support during the competition, please use the Discussion Forum or contact competition organizers directly.

**Good luck, and happy coding!**