Sefika commited on
Commit
487b3ca
·
verified ·
1 Parent(s): 1f468bd

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +55 -24
README.md CHANGED
@@ -4,13 +4,13 @@ configs:
4
  - config_name: benchmark
5
  data_files:
6
  - split: test
7
- path: original_fewrel_inverse.json
8
  task_categories:
9
  - text-classification
10
  - text-generation
11
  language:
12
  - en
13
- pretty_name: FewRel Inverse Relations
14
  size_categories:
15
  - 1K<n<10K
16
  tags:
@@ -18,11 +18,13 @@ tags:
18
  - nlp
19
  - dataset-robustness
20
  - inverse-relations
 
 
21
  ---
22
 
23
  # Reversing Arrows: A Benchmark Dataset for Inverse Relation Directionality in LLMs
24
 
25
- This dataset evaluates the robustness of **relation extraction (RE)** models with a focus on **inverse relations** and **entity substitution**. It is derived from the original **FewRel** dataset and is designed to measure whether models correctly capture relation semantics independent of relation direction and entity identity.
26
 
27
  The dataset includes **MLCommons Croissant metadata** for improved interoperability with machine learning tools and data catalogs.
28
 
@@ -36,21 +38,28 @@ The benchmark investigates whether language models can recognize the same semant
36
  - entity mentions are replaced with synthetic placeholders,
37
  - memorized world knowledge is unavailable.
38
 
39
- The dataset contains two complementary versions:
40
 
41
- ### Original FewRel
42
- - **File:** `original_fewrel_inverse.json`
43
- - Original FewRel sentences with forward and inverse relation annotations.
 
44
 
45
- ### Synthetic FewRel
46
- - **File:** `synthetic_fewrel_inverse.json`
47
- - Entity mentions are replaced with synthetic names (e.g., *Denmark* → *Emilyfort*) to reduce reliance on memorized knowledge.
 
 
48
 
49
  Additionally, the dataset includes:
50
 
51
  ### Relation Metadata
52
- - **File:** `fewrel_inverse_relations.json`
53
- - Human-readable names and definitions for Wikidata relation IDs.
 
 
 
 
54
 
55
  ---
56
 
@@ -67,39 +76,61 @@ Additionally, the dataset includes:
67
 
68
  ## Dataset Structure
69
 
70
- The dataset consists of three JSON files:
 
 
 
 
 
 
71
 
72
  | File | Description |
73
  |------|-------------|
74
- | `original_fewrel_inverse.json` | Original FewRel examples with forward and inverse relation annotations. |
75
- | `synthetic_fewrel_inverse.json` | Entity-substituted version used for robustness evaluation. |
76
- | `fewrel_inverse_relations.json` | Metadata describing Wikidata relation IDs, names, and definitions. |
 
 
77
 
78
  ---
79
 
80
  ## Features
81
 
82
- Each example contains:
83
 
84
  | Field | Description |
85
  |------|-------------|
86
- | `tokens` | Tokenized sentence. |
 
87
  | `head` | Head entity. |
88
  | `tail` | Tail entity. |
89
- | `head_to_tail` | Forward relation. |
90
- | `tail_to_head` | Inverse relation. |
91
- | `artificial_data` | Mapping from original entity names to synthetic replacements (synthetic dataset only). |
 
 
 
 
 
 
 
 
 
 
 
 
92
 
93
  ---
94
 
95
  ## Usage
96
 
97
- Load the dataset using the Hugging Face Datasets library:
98
 
99
  ```python
100
- from datasets import load_dataset
101
 
102
- dataset = load_dataset("Sefika/FewRel_Inverse_Relations")
 
103
  ```
104
 
105
  Potential applications include:
 
4
  - config_name: benchmark
5
  data_files:
6
  - split: test
7
+ path: data/ablation-tekgen/combined_fewrel_tekgen_inverse.json
8
  task_categories:
9
  - text-classification
10
  - text-generation
11
  language:
12
  - en
13
+ pretty_name: Reversing Arrows Inverse Relations Benchmark
14
  size_categories:
15
  - 1K<n<10K
16
  tags:
 
18
  - nlp
19
  - dataset-robustness
20
  - inverse-relations
21
+ - fewrel
22
+ - tekgen
23
  ---
24
 
25
  # Reversing Arrows: A Benchmark Dataset for Inverse Relation Directionality in LLMs
26
 
27
+ This dataset evaluates the robustness of **relation extraction (RE)** models with a focus on **inverse relations** and **entity substitution**. It combines examples from **FewRel** and **TekGen** to measure whether models correctly capture relation semantics independent of relation direction and entity identity.
28
 
29
  The dataset includes **MLCommons Croissant metadata** for improved interoperability with machine learning tools and data catalogs.
30
 
 
38
  - entity mentions are replaced with synthetic placeholders,
39
  - memorized world knowledge is unavailable.
40
 
41
+ The current benchmark release includes a single combined file for evaluation:
42
 
43
+ ### Combined FewRel + TekGen
44
+ - **File:** `data/ablation-tekgen/combined_fewrel_tekgen_inverse.json`
45
+ - **Records:** 5131 (3401 FewRel + 1730 TekGen)
46
+ - Concatenated benchmark file covering both sources.
47
 
48
+ The repository also provides source files used to build the combined file:
49
+
50
+ ### Source Files
51
+ - **FewRel source:** `data/ablation-tekgen/original_fewrel_inverse.json`
52
+ - **TekGen source:** `data/ablation-tekgen/original_tekgen_inverse.json`
53
 
54
  Additionally, the dataset includes:
55
 
56
  ### Relation Metadata
57
+ - **File:** `data/ablation-tekgen/tekgen_relations.json`
58
+ - Human-readable relation and inverse-relation definitions for TekGen relations.
59
+
60
+ ### Croissant Metadata
61
+ - **File:** `data/ablation-tekgen/inverse_relations_croissant.json`
62
+ - MLCommons Croissant description of available resources and fields.
63
 
64
  ---
65
 
 
76
 
77
  ## Dataset Structure
78
 
79
+ Core evaluation file:
80
+
81
+ | File | Description |
82
+ |------|-------------|
83
+ | `data/ablation-tekgen/combined_fewrel_tekgen_inverse.json` | Combined benchmark examples from FewRel and TekGen. |
84
+
85
+ Supporting files:
86
 
87
  | File | Description |
88
  |------|-------------|
89
+ | `data/ablation-tekgen/original_fewrel_inverse.json` | FewRel examples with forward and inverse relation annotations. |
90
+ | `data/ablation-tekgen/original_tekgen_inverse.json` | TekGen examples with forward and inverse relation annotations. |
91
+ | `data/ablation-tekgen/artificial_tekgen_inverse.json` | TekGen entity-substituted examples for robustness analysis. |
92
+ | `data/ablation-tekgen/tekgen_relations.json` | TekGen relation metadata and inverse mapping. |
93
+ | `data/ablation-tekgen/inverse_relations_croissant.json` | Croissant dataset metadata. |
94
 
95
  ---
96
 
97
  ## Features
98
 
99
+ Because the dataset combines two sources, fields vary slightly by source. Common fields are:
100
 
101
  | Field | Description |
102
  |------|-------------|
103
+ | `head_to_tail` | Forward relation label (PID for FewRel or relation string for TekGen). |
104
+ | `tail_to_head` | Inverse relation label. |
105
  | `head` | Head entity. |
106
  | `tail` | Tail entity. |
107
+
108
+ FewRel-specific fields:
109
+
110
+ | Field | Description |
111
+ |------|-------------|
112
+ | `tokens` | Tokenized sentence. |
113
+ | `relation_prop_wiki` | Wikidata relation URIs connected to the entity pair. |
114
+
115
+ TekGen-specific fields:
116
+
117
+ | Field | Description |
118
+ |------|-------------|
119
+ | `id` | Unique example identifier. |
120
+ | `sent` | Full sentence string. |
121
+ | `artificial_data` | Mapping from original to substituted entities (only in artificial TekGen file). |
122
 
123
  ---
124
 
125
  ## Usage
126
 
127
+ Load the combined benchmark file with Python:
128
 
129
  ```python
130
+ import json
131
 
132
+ with open("data/ablation-tekgen/combined_fewrel_tekgen_inverse.json", "r", encoding="utf-8") as f:
133
+ dataset = json.load(f)
134
  ```
135
 
136
  Potential applications include: