File size: 2,531 Bytes
3bb9e3d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
---
license: apache-2.0
language:
- en
task_categories:
- text-classification
tags:
- conversational-ai
- user-satisfaction
- dissatisfaction
- feedback-classification
size_categories:
- 10K<n<100K
---

# User Satisfaction Classification Dataset

A dataset for training classifiers to detect user satisfaction and dissatisfaction reasons from follow-up messages in conversational AI systems.

## Dataset Description

This dataset contains user follow-up messages labeled with satisfaction status and dissatisfaction reasons. The key insight is that **follow-up messages alone contain sufficient signal** for classification—no conversation context is needed.

## Labels

| Label | Description | Count (Train) |
|-------|-------------|---------------|
| `SAT` | User is satisfied | ~3,350 |
| `NEED_CLARIFICATION` | User needs more explanation | ~3,350 |
| `WRONG_ANSWER` | System provided incorrect information | ~3,350 |
| `WANT_DIFFERENT` | User wants alternative options | ~3,350 |

## Dataset Structure

```
train.jsonl          # 13,422 training examples
val.jsonl            # 1,494 validation examples  
adversarial_test.jsonl # 103 adversarial test examples
```

### Data Format

Each line is a JSON object:
```json
{"text": "What do you mean by that?", "label_name": "NEED_CLARIFICATION"}
{"text": "Thanks, that's perfect!", "label_name": "SAT"}
{"text": "No, that's incorrect", "label_name": "WRONG_ANSWER"}
{"text": "Show me other options", "label_name": "WANT_DIFFERENT"}
```

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("rootfs/user-satisfaction-dataset")

# Access splits
train = dataset["train"]
val = dataset["validation"]
test = dataset["test"]

# Example
print(train[0])
# {'text': 'What do you mean?', 'label_name': 'NEED_CLARIFICATION'}
```

## Data Quality

- **Balanced classes**: Equal representation across all 4 labels
- **Normalized text**: 15-100 characters, no format shortcuts
- **100% unique**: No duplicate examples
- **Adversarial test set**: Manually curated challenging examples

## Source

Derived from public conversational AI datasets:
- MIMICS (Microsoft)
- SGD (Google)
- INSCIT

## Associated Model

Pre-trained classifier: [rootfs/dissat-4class](https://huggingface.co/rootfs/dissat-4class)

## Citation

```bibtex
@dataset{user_satisfaction_2024,
  title={User Satisfaction Classification Dataset},
  author={Semantic Router Project},
  year={2024},
  url={https://huggingface.co/datasets/rootfs/user-satisfaction-dataset}
}
```

## License

Apache 2.0