Datasets:
Note the KDD 2027 acceptance and clarify the sources field
Browse files
README.md
CHANGED
|
@@ -57,6 +57,8 @@ configs:
|
|
| 57 |
|
| 58 |
# RetailBridge
|
| 59 |
|
|
|
|
|
|
|
| 60 |
**RetailBridge** is a multi-source question-answering benchmark for enterprise retail.
|
| 61 |
Each question is answered over a heterogeneous retail database with five SQL tables and
|
| 62 |
three free-text stores (reviews, customer inquiries, live-commerce sales scripts).
|
|
@@ -149,7 +151,7 @@ questions through two operators.
|
|
| 149 |
| `answer` | string | Gold answer |
|
| 150 |
| `reasoning_level` | int | 1, 2 or 3 |
|
| 151 |
| `source_chaining` | string | `Homo` or `Hetero` |
|
| 152 |
-
| `sources` | list[string] | Gold evidence sources in hop order
|
| 153 |
| `evidence_modality` | string | `sql-only`, `text-only` or `composite` |
|
| 154 |
| `composite_family` | string or null | `entity_linked` or `answer_linked`; null for Level 1 |
|
| 155 |
| `operation_type` | string | Operation of the final hop (`search`, `aggregate`, `filter`, `rank`, `calculate`, `compare`) |
|
|
@@ -218,7 +220,7 @@ the SQL tables.
|
|
| 218 |
|
| 219 |
The [code repository](https://github.com/DISL-Lab/RetailBridge) contains RetailRAG and
|
| 220 |
the evaluation script. Answer accuracy is judged by an LLM (Muse Ember) against the gold
|
| 221 |
-
answer;
|
| 222 |
[project page](https://retailbridge-kdd.github.io).
|
| 223 |
|
| 224 |
## License
|
|
|
|
| 57 |
|
| 58 |
# RetailBridge
|
| 59 |
|
| 60 |
+
**Accepted at KDD 2027 (Datasets and Benchmarks Track).**
|
| 61 |
+
|
| 62 |
**RetailBridge** is a multi-source question-answering benchmark for enterprise retail.
|
| 63 |
Each question is answered over a heterogeneous retail database with five SQL tables and
|
| 64 |
three free-text stores (reviews, customer inquiries, live-commerce sales scripts).
|
|
|
|
| 151 |
| `answer` | string | Gold answer |
|
| 152 |
| `reasoning_level` | int | 1, 2 or 3 |
|
| 153 |
| `source_chaining` | string | `Homo` or `Hetero` |
|
| 154 |
+
| `sources` | list[string] | Gold evidence sources in hop order (see *Source names*) |
|
| 155 |
| `evidence_modality` | string | `sql-only`, `text-only` or `composite` |
|
| 156 |
| `composite_family` | string or null | `entity_linked` or `answer_linked`; null for Level 1 |
|
| 157 |
| `operation_type` | string | Operation of the final hop (`search`, `aggregate`, `filter`, `rank`, `calculate`, `compare`) |
|
|
|
|
| 220 |
|
| 221 |
The [code repository](https://github.com/DISL-Lab/RetailBridge) contains RetailRAG and
|
| 222 |
the evaluation script. Answer accuracy is judged by an LLM (Muse Ember) against the gold
|
| 223 |
+
answer; results for 16 LLM backbones are on the
|
| 224 |
[project page](https://retailbridge-kdd.github.io).
|
| 225 |
|
| 226 |
## License
|