framolfese commited on
Commit
a5d42f3
·
verified ·
1 Parent(s): 73ef86d

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +86 -0
README.md CHANGED
@@ -47,3 +47,89 @@ configs:
47
  - split: test
48
  path: data/test-*
49
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
47
  - split: test
48
  path: data/test-*
49
  ---
50
+
51
+
52
+
53
+ # 2WikiMultihopQA
54
+
55
+ **This repository only repackages the original 2WikiMultihopQA data so that every example follows the field layout used by [HotpotQA](https://hotpotqa.github.io/).** The content of the underlying questions, answers and contexts is **unaltered**.
56
+ All intellectual credit for creating 2WikiMultihopQA belongs to the authors of the paper *Constructing a Multi‑hop QA Dataset for Comprehensive Evaluation of Reasoning Steps* (COLING 2020) and the accompanying code/data in their GitHub project [https://github.com/Alab-NII/2wikimultihop](https://github.com/Alab-NII/2wikimultihop).
57
+
58
+ ## Dataset Summary
59
+
60
+ * **Name:** 2WikiMultihopQA
61
+ * **What’s different:** only the JSON schema. Each instance now has `id`, `question`, `answer`, `type`, `evidences`, `supporting_facts`, and `context` keys arranged exactly like HotpotQA so that existing HotpotQA data pipelines work out‑of‑the‑box.
62
+
63
+ The dataset still contains **multi‑hop question‑answer pairs** with supporting evidence chains drawn from Wikipedia. There are three splits:
64
+
65
+ | split | #examples |
66
+ | ---------- | --------: |
67
+ | train | 113,284 |
68
+ | validation | 12,981 |
69
+ | test | 12,995 |
70
+
71
+ ## Dataset Structure
72
+
73
+ Each JSON Lines file contains records like:
74
+
75
+ ```json
76
+ {
77
+ "id": "13f5ad2c088c11ebbd6fac1f6bf848b6",
78
+ "question": "Are director of film Move (1970 Film) and director of film Méditerranée (1963 Film) from the same country?",
79
+ "answer": "no",
80
+ "type": "bridge_comparison",
81
+ "level": "unknown",
82
+ "supporting_facts": {
83
+ "title": ["Move (1970 film)", "Méditerranée (1963 film)", "Stuart Rosenberg", "Jean-Daniel Pollet"],
84
+ "sent_id": [0, 0, 0, 0]
85
+ },
86
+ "context": {
87
+ "title": ["Stuart Rosenberg", "Méditerranée (1963 film)", "Move (1970 film)", ...],
88
+ "sentences": [["Stuart Rosenberg (August 11, 1927 – March 15, 2007) was an American film and television director ..."],
89
+ ["Méditerranée is a 1963 French experimental film directed by Jean-Daniel Pollet ..."],
90
+ ...]
91
+ }
92
+ }
93
+ ```
94
+
95
+ ### Field definitions
96
+
97
+ | Field | Type | Description |
98
+ | -------------------------- | -------------------- | ------------------------------------------------------------------------ |
99
+ | `id` | `string` | Unique identifier (original `_id`). |
100
+ | `question` | `string` | Natural‑language question. |
101
+ | `answer` | `string` | Short answer span (may be "yes"/"no" for binary questions). |
102
+ | `type` | `string` | Original 2Wiki question type (e.g. `bridge_comparison`, `comparison`). |
103
+ | `evidences` | `List[List[string]]` | Structured data (subject, property, object) obtained from Wikidata. |
104
+ | `supporting_facts.title` | `List[string]` | Wikipedia page titles that contain evidence sentences. |
105
+ | `supporting_facts.sent_id` | `List[int]` | Zero‑based sentence indices within each page that support the answer. |
106
+ | `context.title` | `List[string]` | Titles for every paragraph provided to the model. |
107
+ | `context.sentences` | `List[List[string]]` | Tokenised sentences for each corresponding title. |
108
+
109
+ ## Data Splits
110
+
111
+ The conversion keeps the same train/validation/test division as the original dataset. No documents or examples were removed or added.
112
+
113
+ ## Source Data
114
+
115
+ * **Original repository:** [https://github.com/Alab-NII/2wikimultihop](https://github.com/Alab-NII/2wikimultihop)
116
+ Contains data generation scripts, the Apache‑2.0 license and citation information.
117
+ * **Paper:** Ho et al., *Constructing A Multi-hop QA Dataset for Comprehensive Evaluation of Reasoning Steps*, COLING 2020.
118
+ [\[ACL Anthology\]](https://aclanthology.org/2020.coling-main.580/) – [\[arXiv\]](https://arxiv.org/abs/2011.01060)
119
+
120
+ No new text was crawled; every paragraph is already present in the original dataset.
121
+
122
+ ## License
123
+
124
+ The original 2WikiMultihopQA is released under the **Apache License 2.0**.
125
+ This redistribution keeps the same license. See [`LICENSE`](./LICENSE) copied verbatim from the upstream repo.
126
+
127
+ ## How to Use
128
+
129
+ ```python
130
+ from datasets import load_dataset
131
+
132
+ # after `ds.push_to_hub(...)` the dataset can be loaded like:
133
+ ds = load_dataset("framolfese/2wiki_multihopqa_hotpotfmt")
134
+ print(ds["train"][0]["question"])
135
+ ```