File size: 1,045 Bytes
4669e6d
6b40fbc
 
 
 
abeea70
4669e6d
6b40fbc
 
 
 
 
 
 
 
 
 
 
abeea70
6b40fbc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
---
tags:
- deduplicated
- semhash
- semantic-deduplication
- hfjobs
---

# Deduplicated imdb

This dataset is a deduplicated version of [imdb](https://huggingface.co/datasets/imdb) 
using semantic deduplication with [SemHash](https://github.com/MinishLab/semhash).

## Deduplication Details

- **Method**: deduplicate
- **Column**: `text`
- **Original size**: 25,000 samples
- **Deduplicated size**: 24,830 samples
- **Duplicate ratio**: 0.68%
- **Reduction**: 0.68%

- **Date processed**: 2025-06-27

## How to use

```python
from datasets import load_dataset

dataset = load_dataset("imdb-deduplicated")
```

## Processing script

This dataset was created using the following script:

```bash
uv run dedupe-dataset.py imdb text <repo_id> --method deduplicate
```

## About semantic deduplication

Unlike exact deduplication, semantic deduplication identifies and removes samples that are 
semantically similar even if they use different words. This helps create cleaner training 
datasets and prevents data leakage between train/test splits.