A collection of Terms of Service or Privacy Policy datasets

Annotated datasets

CUAD

Specifically, the 28 service agreements from CUAD, which are licensed under CC BY 4.0 (subset: cuad).

Code

import datasets
from tos_datasets.proto import DocumentQA

ds = datasets.load_dataset("chenghao/tos_pp_dataset", "cuad")

print(DocumentQA.model_validate_json(ds["document"][0]))

100 ToS

From Annotated 100 ToS, CC BY 4.0 (subset: 100_tos).

Code

import datasets
from tos_datasets.proto import DocumentEUConsumerLawAnnotation

ds = datasets.load_dataset("chenghao/tos_pp_dataset", "100_tos")

print(DocumentEUConsumerLawAnnotation.model_validate_json(ds["document"][0]))

Multilingual Unfair Clause

From CLAUDETTE/Multilingual Unfair Clause, CC BY 4.0 (subset: multilingual_unfair_clause).

It was built from CLAUDETTE/25 Terms of Service in English, Italian, German, and Polish (100 documents in total) from A Corpus for Multilingual Analysis of Online Terms of Service.

Code

import datasets
from tos_datasets.proto import DocumentClassification

ds = datasets.load_dataset("chenghao/tos_pp_dataset", "multilingual_unfair_clause")

print(DocumentClassification.model_validate_json(ds["document"][0]))

Memnet ToS

From 100 Terms of Service in English from Detecting and explaining unfairness in consumer contracts through memory networks, MIT (subset: memnet_tos).

Code

import datasets
from tos_datasets.proto import DocumentClassification

ds = datasets.load_dataset("chenghao/tos_pp_dataset", "memnet_tos")

print(DocumentClassification.model_validate_json(ds["document"][0]))

142 ToS

From 142 Terms of Service in English divided according to market sector from Assessing the Cross-Market Generalization Capability of the CLAUDETTE System, Unknown (subset: 142_tos). This should also includes 50 Terms of Service in English from "CLAUDETTE: an Automated Detector of Potentially Unfair Clauses in Online Terms of Service".

Code

import datasets
from tos_datasets.proto import DocumentClassification

ds = datasets.load_dataset("chenghao/tos_pp_dataset", "142_tos")

print(DocumentClassification.model_validate_json(ds["document"][0]))

10 ToS/PP

From 5 Terms of Service and 5 Privacy Policies in English and German (10 documents in total) from Cross-lingual Annotation Projection in Legal Texts, GNU GPL 3.0 (subset: 10_tos)

Code

import datasets
from tos_datasets.proto import DocumentClassification

ds = datasets.load_dataset("chenghao/tos_pp_dataset", "10_tos")

print(DocumentClassification.model_validate_json(ds["document"][0]))

PolicyQA

This dataset seems to have some annotation issues where unanswerable questions are still answered with SQuAD-v1 format instead of the v2 format.

From PolicyQA, MIT (subset: privacy_glue/policy_qa).

Code

import datasets
from tos_datasets.proto import DocumentQA

ds = datasets.load_dataset("chenghao/tos_pp_dataset", "privacy_glue/policy_qa")

print(DocumentQA.model_validate_json(ds["train"]["document"][0]))

PolicyIE

From PolicyIE, MIT (subset: privacy_glue/policy_ie).

Code

import datasets
from tos_datasets.proto import DocumentSequenceClassification, DocumentEvent

ds = datasets.load_dataset("chenghao/tos_pp_dataset", "privacy_glue/policy_ie")

print(DocumentSequenceClassification.model_validate_json(ds["train"]["type_i"][0]))
print(DocumentEvent.model_validate_json(ds["train"]["type_ii"][0]))

Policy Detection

From [policy-detection-data](https://github.com/infsys-lab/policy-detection-data, GPL 3.0 (subset: privacy_glue/policy_detection).

Code

import datasets
from tos_datasets.proto import DocumentClassification

ds = datasets.load_dataset("chenghao/tos_pp_dataset", "privacy_glue/policy_detection")

print(DocumentClassification.model_validate_json(ds["train"]["document"][0]))

Polisis

From Polisis, Unknown (subset: privacy_glue/polisis).

Code

import datasets
from tos_datasets.proto import DocumentClassification

ds = datasets.load_dataset("chenghao/tos_pp_dataset", "privacy_glue/polisis")

print(DocumentClassification.model_validate_json(ds["test"]["document"][0]))

PrivacyQA

From PrivacyQA, MIT (subset: privacy_qa).

Code

import datasets
from tos_datasets.proto import DocumentClassification

ds = datasets.load_dataset("chenghao/tos_pp_dataset", "privacy_glue/privacy_qa")

print(DocumentClassification.model_validate_json(ds["test"]["document"][0]))

Piextract

From Piextract, Unknown (subset: privacy_glue/piextract).

Code

import datasets
from tos_datasets.proto import DocumentSequenceClassification

ds = datasets.load_dataset("chenghao/tos_pp_dataset", "privacy_glue/piextract")

print(DocumentSequenceClassification.model_validate_json(ds["train"]["document"][0]))

WIP

~~Annotated Italian TOS sentences, Apache 2.0~~ Only sentence level annotations, missing original full text
~~Huggingface, MIT~~ Only sentence level annotations, missing original full text
ToSDR API, Unknown