PatentAgentBench
Collection
PatentAgentBench (PAB) is a benchmark comprising 8 professional patent-related tasks for evaluating LLM agents. • 8 items • Updated
Datasets:
field_values dict | scoring_policy dict | layer_weights dict | metrics listlengths 20 23 | expert_facts unknown |
|---|---|---|---|---|
{
"id": 13,
"apno": "EP2013875593",
"country": "EP",
"applicants": "Halliburton Energy Services, Inc.",
"notify_type": null,
"q_mrdt": "20160629",
"notify_content": "<!-- Page 1 -->\n\n| Datum<br>Date<br>Date | cf Form 1507 | Blatt<br>Sheet<br>Feuille | 1 | Anmelde-Nr:<br>Application No:<br>Demande n°: | ... | {
"automatic_metrics": "RC-1 to RC-5 retain their existing activation conditions, ACC/F1 algorithms, and 0–1 outputs",
"subjective_metrics": "Activated common-layer Judge metrics directly output 0, 0.5, or 1, without field aggregation or GATE zeroing",
"specific_metrics": "Expert MD may optionally configure S ite... | {
"common": 1,
"specific": 0
} | [
{
"metric_id": "RC-1",
"name": "Accuracy of basic information identification",
"kind": "automatic",
"activation_condition": "All samples",
"scoring_method": "Existing ACC",
"fixed_weight": 0.01,
"layer": "common",
"field_refs": [
"#/field_values/apno",
"#/field_values/app... | null |
{"id":27,"apno":"EP2018937981","country":"EP","applicants":"UPM-Kymmene Corporation","notify_type":n(...TRUNCATED) | {"automatic_metrics":"RC-1 to RC-5 retain their existing activation conditions, ACC/F1 algorithms, a(...TRUNCATED) | {
"common": 1,
"specific": 0
} | [{"metric_id":"RC-1","name":"Accuracy of basic information identification","kind":"automatic","activ(...TRUNCATED) | null |
{"id":54,"apno":"EP2011075105","country":"EP","applicants":"New England Biolabs, Inc.","notify_type"(...TRUNCATED) | {"automatic_metrics":"RC-1 to RC-5 retain their existing activation conditions, ACC/F1 algorithms, a(...TRUNCATED) | {
"common": 0.8,
"specific": 0.2
} | [{"metric_id":"RC-1","name":"Accuracy of basic information identification","kind":"automatic","activ(...TRUNCATED) | {"f01":{"fact_id":"f01","text":"本OA主要涉及形式/内容类修改，无需说理论证，因(...TRUNCATED) |
{"id":60,"apno":"EP2011741502","country":"EP","applicants":"Dow Global Technologies LLC","notify_typ(...TRUNCATED) | {"automatic_metrics":"RC-1 to RC-5 retain their existing activation conditions, ACC/F1 algorithms, a(...TRUNCATED) | {
"common": 1,
"specific": 0
} | [{"metric_id":"RC-1","name":"Accuracy of basic information identification","kind":"automatic","activ(...TRUNCATED) | null |
{"id":82,"apno":"EP2017184017","country":"EP","applicants":"Fujitsu Limited","notify_type":null,"q_m(...TRUNCATED) | {"automatic_metrics":"RC-1 to RC-5 retain their existing activation conditions, ACC/F1 algorithms, a(...TRUNCATED) | {
"common": 1,
"specific": 0
} | [{"metric_id":"RC-1","name":"Accuracy of basic information identification","kind":"automatic","activ(...TRUNCATED) | null |
{"id":92,"apno":"EP2017201795","country":"EP","applicants":"Sumitomo Rubber Industries, Ltd.","notif(...TRUNCATED) | {"automatic_metrics":"RC-1 to RC-5 retain their existing activation conditions, ACC/F1 algorithms, a(...TRUNCATED) | {
"common": 1,
"specific": 0
} | [{"metric_id":"RC-1","name":"Accuracy of basic information identification","kind":"automatic","activ(...TRUNCATED) | null |
{"id":93,"apno":"EP2013770071","country":"EP","applicants":"Kyushu University, National University C(...TRUNCATED) | {"automatic_metrics":"RC-1 to RC-5 retain their existing activation conditions, ACC/F1 algorithms, a(...TRUNCATED) | {
"common": 1,
"specific": 0
} | [{"metric_id":"RC-1","name":"Accuracy of basic information identification","kind":"automatic","activ(...TRUNCATED) | null |
{"id":106,"apno":"US15601685","country":"US","applicants":"CASTRO ET AL.","notify_type":"Non-Final R(...TRUNCATED) | {"automatic_metrics":"RC-1 to RC-5 retain their existing activation conditions, ACC/F1 algorithms, a(...TRUNCATED) | {
"common": 1,
"specific": 0
} | [{"metric_id":"RC-1","name":"Accuracy of basic information identification","kind":"automatic","activ(...TRUNCATED) | null |
{"id":120,"apno":"US10569113","country":"US","applicants":"TAKAMATSU ET AL.","notify_type":"Non-Fina(...TRUNCATED) | {"automatic_metrics":"RC-1 to RC-5 retain their existing activation conditions, ACC/F1 algorithms, a(...TRUNCATED) | {
"common": 1,
"specific": 0
} | [{"metric_id":"RC-1","name":"Accuracy of basic information identification","kind":"automatic","activ(...TRUNCATED) | null |
{"id":136,"apno":"US13221391","country":"US","applicants":"MOLLINGER ET AL.","notify_type":"Non-Fina(...TRUNCATED) | {"automatic_metrics":"RC-1 to RC-5 retain their existing activation conditions, ACC/F1 algorithms, a(...TRUNCATED) | {
"common": 1,
"specific": 0
} | [{"metric_id":"RC-1","name":"Accuracy of basic information identification","kind":"automatic","activ(...TRUNCATED) | null |
End of preview. Expand in Data Studio
PatSnap Patent OAR Bench
A benchmark dataset for patent Office Action Response (OAR) tasks. The test split contains 40 cases: 13 CN, 13 EP, and 14 US. It includes 40 Office Action PDF documents totaling 229 pages, structured case materials, expert facts where available, and 812 scoring criteria entries.
- Dataset:
data/test.parquet, 40 cases in Parquet format. - Office Action PDFs:
assets/, namedOAR-<id>.pdfusingfield_values.idfrom each record. - Original source documentation (JSONL layout): English · 中文.
- Source: PatSnap/PatentAgentBench/OAR-bench.
from datasets import load_dataset
from huggingface_hub import hf_hub_download
test = load_dataset("PatSnap/patent-oar-bench", split="test")
case = test[0]
pdf = hf_hub_download(
"PatSnap/patent-oar-bench",
f"assets/OAR-{case['field_values']['id']}.pdf",
repo_type="dataset",
)
The records contain both task material and evaluator reference information. An evaluation setup should define which fields are exposed as model input. The source documentation describes the GitHub JSONL package. On Hugging Face, data/test.parquet is the published test split and assets/ contains the same 40 PDF files.
- Downloads last month
- 107