hackathon-somos-nlp-2023/ask2democracy-cfqa-salud-pension

About Ask2Democracy-cfqa-salud-pension

Ask2Democracy-cfqa-salud-pension is an instructional, context-based generative dataset created using the text reforms of Colombian health and pension systems in Spanish(March 23). The text was pre-processed and augmented using the chat-gpt-turbo API.

Creado por Jorge Henao 🇨🇴 Twitter LinkedIn Linktree
Con el apoyo de David Torres 🇨🇴 Twitter LinkedIn

Different prompt engineering experiments were conducted to obtain high-quality results in Spanish language, while balancing simplicity of language and source-based generation.

Two styles of questions were created:

• Conceptual questions about the text's definitions and concepts related to health and pension reforms.
• First-person questions that address concerns of regular people from diverse population groups.

Conceptual questions and answers are useful for understanding complex concepts commonly found in public documents. These types of questions generally ask for easy-to-understand explanations of a particular concept or subject, such as "What is the definition of concept X or Y?"

On the other hand, first-person questions are designed to address specific concerns of various population groups, including young adults, middle-aged adults, and older adults, as well as those from different gender and labor backgrounds. These questions are tailored to address the specific needs and interests of these groups with regards to the text's topics. These kind of questions looks like this "How x concept can affect me as an y person?, like older adult, labour woman, young adult, etc

Bias

The bias of the text generated by the Chat-GPT-Turbo model has not been evaluated, nor has the accuracy of OpenAI's knowledge about Colombia been verified. Nevertheless, to reduce the risk of model hallucinations and biases, the text was pre-processed, and specific source based contexts were provided on every API call. In addition, source based text was explicity insctucted during the prompt engineering exprimentation for each topic: Health and Pension reforms.

Dataset Structure

Data Instances

plain_text

An example of conceptual questions looks like this.

{
  "instruction": "Given the context please answer the question. Context:'ARTÍCULO 73.  SISTEMA NACIONAL DE PROTECCIÓN SOCIAL...",
  "input": "¿Qué es el Sistema Nacional de Protección Social Integral para la Vejez?",
  "output": "El Sistema Nacional de Protección Social Integral para la Vejez es una iniciativa del gobierno colombiano para garantizar la protección social de las personas mayores. Incluye el Consejo Nacional de Protección a la Vejez y el Comité Técnico, que trabajan para desarrollar políticas y programas que mejoren la calidad de vida de las personas mayores.",
  "topics": ["protección social", "vejez", "políticas públicas"]
},

An example of first person questions looks like this.

{
  "instruction": "Given the context please answer the question. Context:'Artículo 1 Reforma Pensional...' 
  "input": "¿Cómo me beneficia el Sistema Público Único Integrado de Información en Salud (SPUIIS) como adulto de mediana edad?", 
  "output": "El SPUIIS permite el procesamiento digital de imágenes y otras tecnologías de última generación, lo que facilita el diagnóstico y tratamiento de enfermedades, mejorando la calidad de vida de las personas.", 
  "topics": ["Diagnóstico y tratamiento", "Tecnología de última generación", "Calidad de vida"]
}

Data Fields

The data fields are the same among all splits.

plain_text

• instruction: a string feature.
• input: a string feature.
• output: a string feature.