File size: 1,919 Bytes
3c5a6e8 2555730 3c5a6e8 0dd1cc5 2555730 302612e 3c5a6e8 302612e 3c5a6e8 2555730 18f0c87 2555730 18f0c87 2555730 |
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 |
---
license: eupl-1.1
language:
- en
size_categories:
- 100K<n<1M
task_categories:
- feature-extraction
---
European legislation from CELLAR/EUROVOC, English entries only of https://huggingface.co/datasets/EuropeanParliament/Eurovoc.
This data is enriched with embeddings, ready for semantic search.
Last update 16.05.2024: 352011 entries.
## Usage
### With Pandas / Polars
Simply download the parquet file and read with pandas or polars.
```python
import pandas as pd # or import polars as pd
df = pd.read_parquet("CELLAR_EN_16_05_2024.parquet")
df
```
### With HF datsets
```python
from datasets import load_dataset
ds = load_dataset("do-me/Eurovoc_en")
df = ds["train"].to_pandas()
df
```
## Semantic Search
Testwise, the first 512 tokens of every text have been inferenced with the model2vec library and https://huggingface.co/minishlab/M2V_base_output model from @minishlab.
After loading the dataset, use the column `embeddings` for semantic search in this way. See the Jupyter notebook for the full processing script.
You can re-run it on consumer-grade hardware without GPU. Inferencing took `Wall time: 1min 36s` on an M3 Max. Inferecing the entire text takes 50 mins but yields poor quality, currently investigating.
```python
from model2vec import StaticModel
import numpy as np
from sklearn.metrics.pairwise import cosine_similarity
model = StaticModel.from_pretrained("minishlab/M2V_base_output")
query = "social democracy"
quer_emb = model.encode(query)
embeddings_matrix = np.stack(df['embeddings'].to_numpy())
df["cos_sim"] = cosine_similarity(embeddings_matrix, quer_emb.reshape(1, -1))[:, 0]
df = df.sort_values("cos_sim", ascending=False)
df
```
 |