hwaaaaanii commited on
Commit
800076a
·
verified ·
1 Parent(s): ff77331

Improve dataset card readability

Browse files
Files changed (1) hide show
  1. README.md +17 -10
README.md CHANGED
@@ -58,11 +58,17 @@ configs:
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: five SQL tables and
62
  three free-text stores (reviews, customer inquiries, live-commerce sales scripts).
63
- Evidence is spread across these sources, so answering a question means *bridging*
64
- them: identifying the right sources, carrying entities and intermediate answers from
65
- one source into the next, and producing an answer grounded in the retrieved records.
 
 
 
 
 
 
66
 
67
  - **Paper:** [Toward Agentic Retrieval-Augmented Generation for Enterprise Retail Workflows over Heterogeneous Data Sources](https://arxiv.org/abs/2610.25096)
68
  - **Code (RetailRAG evaluation framework):** [github.com/DISL-Lab/RetailBridge](https://github.com/DISL-Lab/RetailBridge)
@@ -117,9 +123,10 @@ Questions are stratified along three evaluation axes.
117
 
118
  Questions were generated by three models (DeepSeek-V4-Pro, Gemini-3.1-Pro, Qwen3.6-Max),
119
  560 each. Level-2 and Level-3 questions are composed from source-grounded single-hop
120
- questions through two operators: **entity-linked** (a later hop is anchored on an entity
121
- identified by an earlier hop) and **answer-linked** (a later hop takes an earlier hop's
122
- answer as a value).
 
123
 
124
  ### Fields
125
 
@@ -130,10 +137,10 @@ answer as a value).
130
  | `answer` | string | Gold answer |
131
  | `reasoning_level` | int | 1, 2 or 3 |
132
  | `source_chaining` | string | `Homo` or `Hetero` |
133
- | `sources` | list[string] | Gold evidence sources, in hop order; the reference for source-routing F1 (see *Source names*) |
134
  | `evidence_modality` | string | `sql-only`, `text-only` or `composite` |
135
  | `composite_family` | string or null | `entity_linked` or `answer_linked`; null for Level 1 |
136
- | `operation_type` | string | Operation of the final hop: `search`, `aggregate`, `filter`, `rank`, `calculate`, `compare` |
137
  | `generator` | string | Model that generated the question |
138
 
139
  ### Example
@@ -157,7 +164,7 @@ answer as a value).
157
 
158
  The database comes from persona-driven simulations of a live-commerce retail platform
159
  (product catalog, broadcast sales scripts, orders, pre- and post-purchase inquiries, and
160
- reviews). Free text lives only in the text stores: the SQL tables carry ids, entities and
161
  structured attributes, so any question that needs what a buyer or seller *said* has to go
162
  through retrieval.
163
 
 
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).
63
+
64
+ Evidence is spread across these sources, so answering a question means *bridging* them.
65
+ A system has to
66
+
67
+ - identify the right sources,
68
+ - carry entities and intermediate answers from one source into the next, and
69
+ - produce an answer grounded in the retrieved records.
70
+
71
+ ## Links
72
 
73
  - **Paper:** [Toward Agentic Retrieval-Augmented Generation for Enterprise Retail Workflows over Heterogeneous Data Sources](https://arxiv.org/abs/2610.25096)
74
  - **Code (RetailRAG evaluation framework):** [github.com/DISL-Lab/RetailBridge](https://github.com/DISL-Lab/RetailBridge)
 
123
 
124
  Questions were generated by three models (DeepSeek-V4-Pro, Gemini-3.1-Pro, Qwen3.6-Max),
125
  560 each. Level-2 and Level-3 questions are composed from source-grounded single-hop
126
+ questions through two operators.
127
+
128
+ - **Entity-linked.** A later hop is anchored on an entity identified by an earlier hop.
129
+ - **Answer-linked.** A later hop takes an earlier hop's answer as a value.
130
 
131
  ### Fields
132
 
 
137
  | `answer` | string | Gold answer |
138
  | `reasoning_level` | int | 1, 2 or 3 |
139
  | `source_chaining` | string | `Homo` or `Hetero` |
140
+ | `sources` | list[string] | Gold evidence sources in hop order, used as the reference for source-routing F1 (see *Source names*) |
141
  | `evidence_modality` | string | `sql-only`, `text-only` or `composite` |
142
  | `composite_family` | string or null | `entity_linked` or `answer_linked`; null for Level 1 |
143
+ | `operation_type` | string | Operation of the final hop (`search`, `aggregate`, `filter`, `rank`, `calculate`, `compare`) |
144
  | `generator` | string | Model that generated the question |
145
 
146
  ### Example
 
164
 
165
  The database comes from persona-driven simulations of a live-commerce retail platform
166
  (product catalog, broadcast sales scripts, orders, pre- and post-purchase inquiries, and
167
+ reviews). Free text lives only in the text stores. The SQL tables carry ids, entities and
168
  structured attributes, so any question that needs what a buyer or seller *said* has to go
169
  through retrieval.
170