datalyes/para_problem

Patent Problem Paraphrase Detection

Dataset Description

This dataset is part of PatenTEB, a comprehensive benchmark for evaluating text embedding models on patent-specific tasks. PatenTEB comprises 15 tasks across retrieval, classification, paraphrase detection, and clustering, with 2.06 million examples designed to reflect real-world patent analysis workflows.

Paper: PatenTEB: A Comprehensive Benchmark and Model Family for Patent Text Embedding

Task Details

• Task Name: para_problem
• Task Type: Sts
• Test Samples: 18,726

Paraphrase detection for patent problem statements. Positive pairs are IN-domain citation pairs where both patents share problem segments; negative pairs are from different domains where semantic overlap is unlikely. Positives comprise 14% of examples.

Dataset Structure

This is a semantic textual similarity (STS) task for paraphrase detection.

Splits:

• test: Text pairs with similarity scores

Columns:

• first_ipcr3
• label
• q
• t
• text1
• text2
• first_ipcr3_count

Data Sample

Below is a 5-row preview of the test set:

first_ipcr3,label,q,t,text1,text2,first_ipcr3_count
A43,0,066-451-545-602-00X,143-409-985-868-372,"provide shoe protectors which protect the insteps of visitors or inspectors at a site where safety shoes are required otherwise, without making the...","provide fluophor, a production method thereof, and a light emitting device",1
A43,0,070-563-486-886-663,100-848-012-270-192,select shoes and/or insole for correcting the posture of foot and judging the biodynamic operating mode of foot,provide a method for guiding a guidance vehicle to a target accurately using a global positioning system by an artificial satellite without relying...,1
A43,0,149-420-479-226-655,007-888-295-884-172,provide a cycle pedal arrangement in which ideal transmission of force is ensured between a shoe and a pedal and the shoe can be positioned optimal...,"provide a buckle, which consists of a buckle main body and an insertion, and the structure of the buckle is simple, and even if a load is applied t...",1
A43,0,081-582-895-069-819,141-117-517-830-130,provide an article of footwear having a ribbed outsole and a midsole with notches,improve dust-collecting capability for dust-like substances contained in a gas and at the same time suppress enlargement of the apparatus size,1
A43,0,060-766-173-693-446,040-714-382-341-674,enable a flap to be optimully and easily opened to easily insert the foot and to maintain sufficient sealability against water or snow at an overla...,provide a method of controlling tint of a tintable window to account for occupant comfort in a room of a building,1

Evaluation Metrics

This task uses Pearson correlation between predicted cosine similarities and binary labels as the primary metric. This measures how well the model's similarity scores align with paraphrase judgments.

Usage

Load Dataset

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("datalyes/{task_name}")

# Access test split
test_data = dataset['test']

Use with Sentence Transformers

from sentence_transformers import SentenceTransformer

# Load a patent-specialized model
model = SentenceTransformer("datalyes/patembed-base")

# Encode patent texts
embeddings = model.encode(test_data['text'])

Integrate with MTEB

This dataset is designed to be integrated with the MTEB (Massive Text Embedding Benchmark) framework. Integration with MTEB is in progress and will be available once the corresponding pull requests are accepted.

Benchmark Context

This dataset is part of a larger benchmark suite:

Benchmark Component	Description
PatenTEB	15 tasks covering retrieval, classification, paraphrase, clustering
Test Data (Released)	319,320 examples across all 15 tasks
Training/Validation Data	1.74 million examples (planned for future release)
Total Dataset Size	2.06 million annotated instances

Note: Currently, only the test split is publicly available. Training and validation data release is planned for a future date.

All 15 Tasks (NEW to MTEB):

• 3 classification tasks: Bloom timing, NLI directionality, IPC3 classification
• 2 clustering tasks: IPC-based, Inventor-based
• 8 retrieval tasks: 3 symmetric (IN/MIXED/OUT domain) + 5 asymmetric (fragment-to-full)
• 2 paraphrase tasks: Problem and solution paraphrase detection

MTEB Integration: Upcoming (PR in progress)

Citation

If you use this dataset, please cite our paper:

@misc{ayaou2025patentebcomprehensivebenchmarkmodel,
      title={PatenTEB: A Comprehensive Benchmark and Model Family for Patent Text Embedding}, 
      author={Iliass Ayaou and Denis Cavallucci},
      year={2025},
      eprint={2510.22264},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2510.22264}
}

License

This dataset is released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license.

• You are free to share and adapt the material
• You must give appropriate credit
• You may not use the material for commercial purposes
• If you remix, transform, or build upon the material, you must distribute your contributions under the same license

For full license details, see: https://creativecommons.org/licenses/by-nc-sa/4.0/

Contact

• Authors: Iliass Ayaou, Denis Cavallucci
• Institution: ICUBE Laboratory, INSA Strasbourg
• GitHub: github.com/iliass-y/patenteb
• HuggingFace: huggingface.co/datalyes