RafaMann/DIMMI

Dataset Card for Dataset Name

DIMMI - Drug InforMation Mining in Italian

Dataset Details

Dataset Description

DIMMI consists of 600 Italian drug package leaflets. The documents in the DIMMI exhibit a wide range of lengths, with the shortest document containing 363 tokens and the longest extending to 11,730 tokens.

• Curated by: Raffaele Manna, Maria Pia di Buono, Luca Giordano -- Language(s) (NLP): Italian
• License: [More Information Needed]

Dataset Sources

DIMMI dataset is derived from the D-LeafIT Corpus, made up of 1819 Italian drug package leaflets. The corpus has been created extracting PILs available on the Italian Agency for Medications (Agenzia Italiana del Farmaco - AIFA), among which 1439 refer to generic drugs and 380 to class A drugs.

• Repository: D-LeafIT - Drug Leaflet Italian Corpus
• Paper: Large Language Models as Drug Information Providers for Patients
• Demo [optional]: [More Information Needed]

Uses

The primary objective of this dataset is to facilitate the development of natural language processing models in the pharmaceutical domain.

Specifically, it is designed to support:

1. Information Extraction Tasks: The dataset serves as a foundation for training models to extract critical information from pharmaceutical texts.

2. Question Answering Systems: It enables the creation of models capable of accurately answering specific questions related to:

   • Drug dosage
   • Medication usage
   • Side effects
   • Drug-drug interactions

Direct Use

Model Evaluation: Researchers and developers can use this dataset to benchmark the performance of their NLP models in pharmaceutical information extraction tasks.

AI in Healthcare: The dataset contributes to the broader goal of enhancing AI applications in healthcare, particularly in pharmacology and patient care.

By approaching this challenge as an information extraction task, researchers can develop models that not only understand pharmaceutical texts but also extract and structure relevant information, making it readily accessible for healthcare professionals and patients alike.

[More Information Needed]

Out-of-Scope Use

While this dataset is designed to support research and development in pharmaceutical information extraction, it's important to note its limitations and potential misuses.

The following uses are considered out of scope and should be avoided:

1. Medical Diagnosis or Treatment: This dataset is not intended to be used as a substitute for professional medical advice, diagnosis, or treatment. It should not be used to make clinical decisions without consultation with qualified healthcare professionals.

2. Drug Prescription: The information extracted from this dataset should not be used to prescribe, modify, or discontinue any medication regimen without proper medical supervision.

3. Legal or Regulatory Compliance: The dataset is not designed to ensure compliance with pharmaceutical regulations or legal requirements. It should not be used as a sole source for regulatory or legal decision-making in the pharmaceutical industry.

4. Patient-Specific Recommendations: The dataset does not account for individual patient characteristics, medical history, or specific health conditions. It should not be used to generate personalized medical recommendations.

5. Commercial Product Development: While the dataset can inform research, it is not intended for direct commercial application in drug development or marketing without additional validation and regulatory approval.

6. Comprehensive Drug Information: The dataset does not cover all existing drugs or all possible drug interactions. It should not be considered an exhaustive source of pharmaceutical information.

Users of this dataset should be aware of these limitations and ensure that their applications align with ethical guidelines and professional standards in healthcare and pharmaceutical research.

Dataset Structure

The DIMMI dataset consists of two main files:

1. dimmi_600.tsv This file contains the primary data entries, where each entry includes:

   • ID: A unique identifier for each entry
   • ID_LOC: Indicates the ID location in the original corpus
   • Drug_Name: The name of the drug
   • Text: The full text of the drug leaflet

2. dimmi_gs_07.json This file contains the gold standard annotations, structured as follows:

   • ID_LOC: Corresponds to the ID_LOC in dimmi_600.tsv
   • Annotations for the following categories, each represented as a list of strings:
     • Molecule
     • Usage
     • Dosage
     • Drug Interaction
     • Side Effect

The use of lists for the annotations in dimmi_gs_07.json allows for multiple entries per category, providing a comprehensive and flexible representation of the annotated information.

Dataset Creation

Curation Rationale

[More Information Needed]

Source Data

Data Collection and Processing

[More Information Needed]

Who are the source data producers?

Annotations

We developed a comprehensive gold standard (GS). This GS was created through manual annotation of the following key categories:

1. Molecule
2. Dosage
3. Drug Interaction
4. Usage
5. Side Effect

The manual annotation process ensures a high-quality benchmark against which we can evaluate our system's accuracy and effectiveness in identifying and categorizing pharmaceutical information.

Annotation process

[More Information Needed]

Who are the annotators?

[More Information Needed]

Personal and Sensitive Information

[More Information Needed]

Bias, Risks, and Limitations

[More Information Needed]

Recommendations

Users should be made aware of the risks, biases and limitations of the dataset. More information needed for further recommendations.

Citation

BibTeX:

@inproceedings{manna2024dimmi, title={DIMMI-Drug InforMation Mining in Italian: A CALAMITA Challenge}, author={Manna, Raffaele and di Buono, Maria Pia and Giordano, Luca and others}, booktitle={Proceedings of the 10th Italian Conference on Computational Linguistics (CLiC-it 2024)}, year={2024} }

APA:

Manna, R., di Buono, M. P., & Giordano, L. (2024). DIMMI-Drug InforMation Mining in Italian: A CALAMITA Challenge. In Proceedings of the 10th Italian Conference on Computational Linguistics (CLiC-it 2024).