moriyad commited on
Commit
b6c5070
·
verified ·
1 Parent(s): 86f1c23

Initial release

Browse files
Files changed (2) hide show
  1. README.md +81 -0
  2. data/validation.parquet +3 -0
README.md ADDED
@@ -0,0 +1,81 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-nc-nd-4.0
3
+ language:
4
+ - en
5
+ pretty_name: ContractGraphEval
6
+ size_categories:
7
+ - n<1K
8
+ task_categories:
9
+ - text-generation
10
+ tags:
11
+ - legal
12
+ - contracts
13
+ - knowledge-graph
14
+ - evaluation
15
+ - llm-as-a-judge
16
+ - uncertainty
17
+ configs:
18
+ - config_name: default
19
+ data_files:
20
+ - split: validation
21
+ path: data/validation.parquet
22
+ ---
23
+
24
+ # ContractGraphEval
25
+
26
+ 150 contract clauses, each with a reference graph, a model-predicted graph, scores from several judges, and uncertainty estimates. Benchmark of **"LLM as a Judge for Evaluating Contract Graphs"** (*Procedia Computer Science* 275, 2026).
27
+
28
+ [Paper](https://doi.org/10.1016/j.procs.2026.01.060) · [Code](https://github.com/moriyadechtiar/ensemble-graph-eval) · [Source dataset](https://huggingface.co/datasets/moriyad/contract-clause-graphs)
29
+
30
+ ## Use
31
+
32
+ ```python
33
+ from datasets import load_dataset
34
+
35
+ ds = load_dataset("moriyad/ContractGraphEval", split="validation")
36
+ s = ds[0]
37
+ s["clause"]["text"], s["judge_scores"]["hybrid"], s["error_prediction"]["p_error_calibrated"]
38
+ ```
39
+
40
+ ## Fields
41
+
42
+ | Field | Description |
43
+ |---|---|
44
+ | `clause`, `instruction` | The clause and the extraction prompt |
45
+ | `gold_graph`, `prediction` | Reference graph and the output of the GRPO-trained Llama-3.1-8B model |
46
+ | `judge_scores` | 8 reference-based scores in [0, 1] |
47
+ | `reference_free_judges` | 7 checks that need no reference graph, computed with the [`contract_linter`](https://github.com/moriyadechtiar/graph-grpo-lex) package |
48
+ | `uncertainty`, `risk_scores`, `error_prediction` | Judge disagreement, per-judge risk, and calibrated error probability |
49
+ | `evaluation` | `has_error`, `mean_score`, `structural_f1` |
50
+ | `source_id` | The same clause in the source dataset |
51
+
52
+ ## Reported results
53
+
54
+ | Result | Value | Error label |
55
+ |---|---|---|
56
+ | Error-detection AUC of judge disagreement (`uncertainty.cv`) | 0.982 | any judge below 0.5 (40 of 150) |
57
+ | Calibration error of `p_error_calibrated` | 0.027 | hybrid score below 0.60 (11 of 150) |
58
+ | Reviewing the 15 highest-risk samples | catches 10 of 11 errors | hybrid score below 0.60 |
59
+
60
+ Error labels are derived from judge scores against the reference graph.
61
+
62
+ ## Notes
63
+
64
+ - The samples are validation rows of the source dataset, from 3 contracts that were not used for training.
65
+ - Version 1.1.
66
+
67
+ ## Licence and citation
68
+
69
+ `cc-by-nc-nd-4.0`. Clause text comes from CUAD (The Atticus Project, CC BY 4.0).
70
+
71
+ ```bibtex
72
+ @article{dechtiar2026llmjudge,
73
+ title = {LLM as a Judge for Evaluating Contract Graphs: Multi-Judge Benchmarking and Agentic Uncertainty-Aware Refinement},
74
+ author = {Dechtiar, Moriya and Katz, Daniel Martin and Jaume, Sylvain and Wang, Hongming},
75
+ journal = {Procedia Computer Science},
76
+ volume = {275},
77
+ pages = {513--522},
78
+ year = {2026},
79
+ doi = {10.1016/j.procs.2026.01.060}
80
+ }
81
+ ```
data/validation.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6dbbfa7e63279ea71c9ce824508f6974e1d15fdfe0856506eebc2aa52b20265d
3
+ size 150732