Update README.md
Browse files
README.md
CHANGED
|
@@ -9,6 +9,8 @@ tags:
|
|
| 9 |
configs:
|
| 10 |
- config_name: queries
|
| 11 |
data_files:
|
|
|
|
|
|
|
| 12 |
- split: test
|
| 13 |
path: queries/test.parquet
|
| 14 |
- config_name: corpus
|
|
@@ -17,43 +19,83 @@ configs:
|
|
| 17 |
path: corpus/corpus.parquet
|
| 18 |
- config_name: qrels
|
| 19 |
data_files:
|
|
|
|
|
|
|
| 20 |
- split: test
|
| 21 |
path: qrels/test.parquet
|
| 22 |
---
|
| 23 |
|
| 24 |
# skill-extraction-techwolf-graded
|
| 25 |
|
| 26 |
-
Graded-relevance
|
| 27 |
[`TechWolf/skill-extraction-techwolf`](https://huggingface.co/datasets/TechWolf/skill-extraction-techwolf)
|
| 28 |
against the ESCO v1.1.0 skill taxonomy. Layout follows the
|
| 29 |
-
[BEIR](https://github.com/beir-cellar/beir) convention
|
| 30 |
-
|
| 31 |
-
|
|
|
|
| 32 |
|
| 33 |
## Configs
|
| 34 |
|
| 35 |
-
| config | rows | columns |
|
| 36 |
-
|---|---:|---|
|
| 37 |
-
| `queries` |
|
| 38 |
-
| `
|
| 39 |
-
| `
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 40 |
|
| 41 |
-
##
|
| 42 |
|
| 43 |
-
|
| 44 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 45 |
|
| 46 |
-
The test split is **not yet fully graded** (0-4). Its relevance labels are
|
| 47 |
-
**real but binary**: `score = 1` marks a genuinely relevant target (derived from
|
| 48 |
-
the public non-graded ground truths), and every pair not listed is implicit
|
| 49 |
-
grade 0. The fine-grained 0-4 graded annotations are withheld during the ongoing
|
| 50 |
-
RecSys-HR challenge (see [`WorkRB website`](https://techwolf-ai.github.io/workrb-site/challenges/recsys-hr-2026.html)) and will be released afterwards.
|
| 51 |
|
| 52 |
## Attribution
|
| 53 |
|
| 54 |
This dataset uses the ESCO classification of the European Commission
|
| 55 |
(ESCO v1.1.0, <https://esco.ec.europa.eu>), licensed under
|
| 56 |
-
[CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). The ESCO content
|
| 57 |
-
|
| 58 |
English description); no semantic modifications were made. The European
|
| 59 |
Commission is not responsible for any use of the data.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
configs:
|
| 10 |
- config_name: queries
|
| 11 |
data_files:
|
| 12 |
+
- split: validation
|
| 13 |
+
path: queries/validation.parquet
|
| 14 |
- split: test
|
| 15 |
path: queries/test.parquet
|
| 16 |
- config_name: corpus
|
|
|
|
| 19 |
path: corpus/corpus.parquet
|
| 20 |
- config_name: qrels
|
| 21 |
data_files:
|
| 22 |
+
- split: validation
|
| 23 |
+
path: qrels/validation.parquet
|
| 24 |
- split: test
|
| 25 |
path: qrels/test.parquet
|
| 26 |
---
|
| 27 |
|
| 28 |
# skill-extraction-techwolf-graded
|
| 29 |
|
| 30 |
+
Graded-relevance annotations for sentences from
|
| 31 |
[`TechWolf/skill-extraction-techwolf`](https://huggingface.co/datasets/TechWolf/skill-extraction-techwolf)
|
| 32 |
against the ESCO v1.1.0 skill taxonomy. Layout follows the
|
| 33 |
+
[BEIR](https://github.com/beir-cellar/beir) convention so it is drop-in for
|
| 34 |
+
MTEB-style retrieval evaluators.
|
| 35 |
+
|
| 36 |
+
This dataset was created for the RecSys-HR 2026 WorkRB challenge.
|
| 37 |
|
| 38 |
## Configs
|
| 39 |
|
| 40 |
+
| config | split | rows | columns |
|
| 41 |
+
|---|---|---:|---|
|
| 42 |
+
| `queries` | `validation` | 75 | `_id` (sentence id), `text` (sentence) |
|
| 43 |
+
| `queries` | `test` | 324 | `_id` (sentence id), `text` (sentence) |
|
| 44 |
+
| `corpus` | `corpus` | 13,891 | `_id` (ESCO skill URI), `title` (English preferred label), `text` (English description), `esco_version` |
|
| 45 |
+
| `qrels` | `validation` | 1,041,825 | `query-id`, `corpus-id`, `score` (0-4) |
|
| 46 |
+
| `qrels` | `test` | 4,500,684 | `query-id`, `corpus-id`, `score` (0-4) |
|
| 47 |
+
|
| 48 |
+
Higher is more relevant. **The corpus is the COMPLETE ESCO v1.1.0 skill
|
| 49 |
+
list (13,891 skills)** The qrels are the full query x corpus grid:
|
| 50 |
+
every query is scored against all 13,891 skills, with score 0 for any skill
|
| 51 |
+
that was not surfaced as relevant.
|
| 52 |
+
|
| 53 |
+
### Score scale, `validation` split
|
| 54 |
+
|
| 55 |
+
| score | volume | meaning |
|
| 56 |
+
|---:|---:|---|
|
| 57 |
+
| 0 | 1,000,838 | The skill is totally unrelated to the sentence. |
|
| 58 |
+
| 1 | 40,060 | The skill's domain is correct. It's a plausible skill in a broader context, but not mentioned in this sentence. |
|
| 59 |
+
| 2 | 598 | The skill could be recommended, but its granularity makes it not core to the query. |
|
| 60 |
+
| 3 | 154 | The skill is strongly relevant for this query, although it is more implied than explicitly demonstrated. |
|
| 61 |
+
| 4 | 175 | The skill is explicitly demonstrated or requested by the query, and is therefore a clearly correct recommendation. |
|
| 62 |
+
|
| 63 |
+
### Score scale, `test` split
|
| 64 |
+
|
| 65 |
+
| score | volume | meaning |
|
| 66 |
+
|---:|---:|---|
|
| 67 |
+
| 0 | 4,280,810 | The skill is totally unrelated to the sentence. |
|
| 68 |
+
| 1 | 214,208 | The skill's domain is correct. It's a plausible skill in a broader context, but not mentioned in this sentence. |
|
| 69 |
+
| 2 | 3,197 | The skill could be recommended, but its granularity makes it not core to the query. |
|
| 70 |
+
| 3 | 1,720 | The skill is strongly relevant for this query, although it is more implied than explicitly demonstrated. |
|
| 71 |
+
| 4 | 749 | The skill is explicitly demonstrated or requested by the query, and is therefore a clearly correct recommendation. |
|
| 72 |
|
| 73 |
+
## Usage
|
| 74 |
|
| 75 |
+
```python
|
| 76 |
+
from datasets import load_dataset
|
| 77 |
+
|
| 78 |
+
queries = load_dataset("TechWolf/Skill-extraction-TechWolf-graded", "queries", split="test")
|
| 79 |
+
corpus = load_dataset("TechWolf/Skill-extraction-TechWolf-graded", "corpus", split="corpus")
|
| 80 |
+
qrels = load_dataset("TechWolf/Skill-extraction-TechWolf-graded", "qrels", split="test")
|
| 81 |
+
```
|
| 82 |
+
|
| 83 |
+
Join on ids: `qrels["query-id"]` matches `queries["_id"]`, and
|
| 84 |
+
`qrels["corpus-id"]` matches `corpus["_id"]` (a full ESCO skill URI).
|
| 85 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 86 |
|
| 87 |
## Attribution
|
| 88 |
|
| 89 |
This dataset uses the ESCO classification of the European Commission
|
| 90 |
(ESCO v1.1.0, <https://esco.ec.europa.eu>), licensed under
|
| 91 |
+
[CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). The ESCO content has
|
| 92 |
+
been extracted into a tabular subset (skill URI, English preferred label,
|
| 93 |
English description); no semantic modifications were made. The European
|
| 94 |
Commission is not responsible for any use of the data.
|
| 95 |
+
|
| 96 |
+
Source sentences come from
|
| 97 |
+
[`TechWolf/skill-extraction-techwolf`](https://huggingface.co/datasets/TechWolf/skill-extraction-techwolf)
|
| 98 |
+
(also CC BY 4.0). The judge labels themselves are released under CC BY 4.0.
|
| 99 |
+
|
| 100 |
+
## Citation
|
| 101 |
+
To be announced
|