metadata

language:
  - en
dataset_info:
  features:
    - name: title
      dtype: string
    - name: content
      dtype: string
    - name: response
      dtype: string
    - name: likes
      dtype: float64
    - name: Section
      dtype: string
    - name: instruct
      dtype: string
  splits:
    - name: train
      num_bytes: 104247254
      num_examples: 61543
  download_size: 65330722
  dataset_size: 104247254
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
task_categories:
  - question-answering
  - text-classification
  - summarization
size_categories:
  - 10K<n<100K

Dataset Card for Llama3-Naija_v1

This Dataset is a Question (Post) - Answer (Response) dataset webscraped from Nairaland. The Dataset involves Questions on various sections from Nairaland ("Politics","Romance","Career","Business","Education","Religion","Sports","Literature","Fashion","TV-Movies","Travel","Programming","Phones","Music-Radio","Food","Family","Health") and the most liked response to those posts.

The data was built with the aim of training or fine-tuning an LLM to chat like a Nigerian.

Dataset Details

Dataset Description

This dataset contains questions and answers from Nairaland, a popular Nigerian online community. The questions cover various topics relevant to Nigerians, and the answers are the most liked responses to those questions. The dataset aims to capture the linguistic and cultural nuances of Nigerian English and Pidgin English, making it suitable for training language models that understand and generate text in these languages.

Developed by: Saheedniyi
Language(s): English, Pidgin English

Dataset Sources

Nairaland.com: Africa's largest internet Forum
GitHub Repository

Uses

Direct Use

This dataset is intended for training or fine-tuning language models to generate or understand Nigerian English and Pidgin English. Suitable use cases include:

Chatbots tailored for Nigerian audiences
Sentiment analysis specific to Nigerian contexts
Cultural and context-aware text generation

Out-of-Scope Use

This dataset is not suitable for:

Generating non-Nigerian context-specific content
Highly sensitive applications without proper ethical review

Dataset Structure

The dataset consists of two main fields for each entry:

title: The title or headline of the original post or question on Nairaland. It provides a quick summary or the main topic of the post.
content: The detailed content of the original post or question. This includes the full text that describes the user's query or discussion point.
response: The most liked response to the original post. This is the answer or reply that received the most positive engagement from other users.
likes: The number of likes the response received. This indicates how many users found the response helpful or relevant.
section: The specific Nairaland section or category where the post was made, such as "Politics," "Romance," "Career," etc. This helps categorize the post by topic.

Dataset Creation

Curation Rationale

The dataset was created to provide a resource for developing AI models that can understand and generate text in a way that is culturally and contextually relevant to Nigerians. This helps in building more effective and relatable AI applications for Nigerian users.

Source Data

Data Collection and Processing

Data was collected by web scraping Nairaland, focusing on the most liked responses to posts across various sections. The data was then cleaned to remove any HTML tags, advertisements, and irrelevant content. Tools and libraries used include Python, BeautifulSoup, and pandas.

Who are the source data producers?

The source data producers are the users of Nairaland who create posts and responses. Demographic information about these users is not collected, but they represent a wide range of Nigerian society.

Annotations

Annotation process

No additional annotations were made beyond the original likes on responses. The most liked response was automatically selected as the answer.

Who are the annotators?

The annotators are the users of Nairaland whose likes determine the most relevant responses.

Personal and Sensitive Information

The dataset does not include any personal identifiable information (PII). Usernames and other identifiers have been anonymized or removed to protect privacy.

Bias, Risks, and Limitations

The dataset reflects the views and biases of the Nairaland community, which may not be representative of the entire Nigerian population. Users should be cautious of these biases, especially for applications that require high accuracy and fairness.

Recommendations

Users should:

Be aware of potential biases and limitations.
Consider the ethical implications of their applications.

Citation

BibTeX:

@dataset{saheedniyi2024llama3naija,
  author       = {Azeez Saheed},
  title        = {Llama3-Naija_v1: A Nairaland Question-Answer Dataset},
  year         = 2024,
  url          = {https://github.com/saheedniyi02/Llama3-8b-Naija_v1},
}

APA:

Azeez Saheed. (2024). Llama3-Naija_v1: A Nairaland Question-Answer Dataset. Retrieved from https://github.com/saheedniyi02/Llama3-8b-Naija_v1

Glossary

LLM: Large Language Model
Nairaland: A popular Nigerian online community
Pidgin English: A creole language spoken across Nigeria

More Information

For more details and updates, visit the GitHub Repository.

Dataset Card Authors

This dataset card was prepared by Azeez Saheed.

Dataset Card Contact

For any questions or issues, please contact Azeez Saheed.