File size: 6,767 Bytes
b58a9fb
 
 
 
 
 
 
 
 
d369a31
 
 
cfc216f
b58a9fb
3ec4f44
b58a9fb
cfc216f
b58a9fb
cfc216f
b58a9fb
 
 
 
 
 
 
 
a5e0fdc
 
d2ac645
a5e0fdc
 
 
 
 
e161a37
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b58a9fb
 
 
 
 
 
 
 
cfc216f
 
 
 
b58a9fb
cfc216f
b58a9fb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
23ab49d
b58a9fb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c279948
9f78956
b58a9fb
 
c279948
 
 
 
 
 
 
 
 
 
b58a9fb
 
c279948
b58a9fb
 
 
 
 
 
 
 
c279948
b58a9fb
 
 
 
 
 
 
 
 
 
 
c279948
b58a9fb
 
 
 
 
 
c279948
b58a9fb
 
 
 
 
 
c279948
23ab49d
b58a9fb
 
 
 
 
 
 
c279948
b58a9fb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c279948
b58a9fb
 
 
cfc216f
b58a9fb
 
 
 
 
 
 
 
 
 
 
 
 
8647bad
 
b58a9fb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
23ab49d
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
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
---
license: mit
language:
- en
tags:
- conversations
- tagging
- embeddings
- bittensor
- dialog
- social media
- podcast
pretty_name: 5,000 Podcast Conversations with Metadata and Embedding Dataset
size_categories:
- 1M<n<10M
---
## πŸ—‚οΈ ReadyAI - 5,000 Podcast Conversations with Metadata and Embedding Dataset

ReadyAI, operating subnet 33 on the [Bittensor Network](https://bittensor.com/) is an open-source initiative focused on low-cost, resource-minimal pipelines for structuring raw data for AI applications.
This dataset is part of the ReadyAI Conversational Genome Project, leveraging the Bittensor decentralized network.

AI runs on structured data β€” and this dataset bridges the gap between raw conversation transcripts and structured, vectorized semantic tags.

You can find more about our subnet on GitHub [here](https://github.com/afterpartyai/bittensor-conversation-genome-project).

---

## Full Vectors Access

➑️ **Download the full 45 GB conversation tags embeddings** from [here](https://huggingface.co/datasets/ReadyAi/5000-podcast-conversations-with-metadata-and-embedding-dataset/tree/main/data)

For large-scale processing and fine-tuning.

---

## πŸ“¦ Dataset Versions

In addition to the full dataset, two smaller versions are available:

- **Small version**
  - Located in the `small_dataset` folder.
  - Contains 1,000 conversations with the same file structure as the full dataset.
  - All filenames are prefixed with `small_`.

- **Medium version**
  - Located in the `medium_dataset` folder.
  - Contains 2,500 conversations
  - Also using the same structure and `medium_` prefix for all files.

These subsets are ideal for lightweight experimentation, prototyping, or benchmarking.

---

## πŸ“‹ Dataset Overview

This dataset contains **annotated conversation transcripts** with:
- Human-readable semantic tags
- **Embedding vectors** contextualized to each conversation
- Participant metadata

It is ideal for:
- Semantic search over conversations
- AI assistant training (OpenAI models, fine-tuning)
- Vector search implementations using **pg_vector** and **Pinecone**
- Metadata analysis and tag retrieval for LLMs

The embeddings were generated with the [text-embedding-ada-002](https://huggingface.co/Xenova/text-embedding-ada-002) model and have 1536 dimensions per tag.
---

## πŸ“‚ Dataset Structure

The dataset consists of four main components:

### 1. **data/bittensor-conversational-tags-and-embeddings-part-*.parquet** β€” Tag Embeddings and Metadata

Each Parquet file contains rows with:


| Column | Type | Description |
|:-------|:-----|:------------|
| c_guid | int64 | Unique conversation group ID |
| tag_id | int64 | Unique identifier for the tag |
| tag | string | Semantic tag (e.g., "climate change") |
| vector | list of float32 | Embedding vector representing the tag's meaning **in the conversation's context** |


βœ… Files split into ~1GB chunks for efficient loading and streaming.

---

### 2. **tag_to_id.parquet** β€” Tag Mapping

Mapping between tag IDs and human-readable tags.

| Column | Type | Description |
|:-------|:-----|:------------|
| tag_id | int64 | Unique tag ID |
| tag | string | Semantic tag text |

βœ… Useful for reverse-mapping tags from models or outputs.

---

### 3. **conversations_to_tags.parquet** β€” Conversation-to-Tag Mappings

Links conversations to their associated semantic tags.

| Column | Type | Description |
|:-------|:-----|:------------|
| c_guid | int64 | Conversation group ID |
| tag_ids | list of int64 | List of tag IDs relevant to the conversation |

βœ… For supervised training, retrieval tasks, or semantic labeling.

---

### 4. **conversations_train.parquet** β€” Full Conversation Text and Participants

Contains the raw multi-turn dialogue and metadata.

| Column | Type | Description |
|:-------|:-----|:------------|
| c_guid | int64 | Conversation group ID |
| transcript | string | Full conversation text |
| participants | list of strings | List of speaker identifiers |

βœ… Useful for dialogue modeling, multi-speaker AI, or fine-tuning.

---

## πŸš€ How to Use

**Install dependencies**

```python
pip install pandas pyarrow datasets
```

**Download the dataset**
```python
import datasets

path = "ReadyAi/5000-podcast-conversations-with-metadata-and-embedding-dataset"
dataset = datasets.load_dataset(path)

print(dataset['train'].column_names)
```

**Load a single Parquet split**

```python
import pandas as pd

df = pd.read_parquet("data/bittensor-conversational-tags-and-embeddings-part-0000.parquet")
print(df.head())
```

**Load all tag splits**

```python
import pandas as pd
import glob

files = sorted(glob.glob("data/bittensor-conversational-tags-and-embeddings-part-*.parquet"))
df_tags = pd.concat((pd.read_parquet(f) for f in files), ignore_index=True)

print(f"Loaded {len(df_tags)} tag records.")
```

**Load tag dictionary**

```python
tag_dict = pd.read_parquet("tag_to_id.parquet")
print(tag_dict.head())
```

**Load conversation to tags mapping**

```python
df_mapping = pd.read_parquet("conversations_to_tags.parquet")
print(df_mapping.head())
```

**Load full conversations dialog and metadata**

```python
df_conversations = pd.read_parquet("conversations_train.parquet")
print(df_conversations.head())
```

---

## πŸ”₯ Example: Reconstruct Tags for a Conversation

```python
# Build tag lookup
tag_lookup = dict(zip(tag_dict['tag_id'], tag_dict['tag']))

# Pick a conversation
sample = df_mapping.iloc[0]
c_guid = sample['c_guid']
tag_ids = sample['tag_ids']

# Translate tag IDs to human-readable tags
tags = [tag_lookup.get(tid, "Unknown") for tid in tag_ids]

print(f"Conversation {c_guid} has tags: {tags}")
```

---

## πŸ“¦ Handling Split Files

| Situation | Strategy |
|:----------|:---------|
| Enough RAM | Use `pd.concat()` to merge splits |
| Low memory | Process each split one-by-one |
| Hugging Face datasets | Use streaming mode |

**Example (streaming with Hugging Face `datasets`)**

```python
from datasets import load_dataset

dataset = load_dataset(
    "ReadyAi/5000-podcast-conversations-with-metadata-and-embedding-dataset",
    split="train",
    streaming=True
)

for example in dataset:
    print(example)
    break
```

---

## πŸ“œ License

MIT License  
βœ… Free to use and modify

---

## ✨ Credits

Built using contributions from Bittensor conversational miners and the ReadyAI open-source community.

---

## 🎯 Summary

| Component | Description |
|:----------|:------------|
| parquets/part_*.parquet | Semantic tags and their contextual embeddings |
| tag_to_id.parquet | Dictionary mapping of tag IDs to text |
| conversations_to_tags.parquet | Links conversations to tags |
| conversations_train.parquet | Full multi-turn dialogue with participant metadata |