---
dataset_info:
  features:
  - name: query
    dtype: string
  - name: positive
    list: string
  - name: negative
    list: string
  splits:
  - name: train
    num_bytes: 74263739
    num_examples: 12724
  - name: dev
    num_bytes: 238604
    num_examples: 200
  - name: test
    num_bytes: 239279
    num_examples: 200
  download_size: 42752991
  dataset_size: 74741622
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: dev
    path: data/dev-*
  - split: test
    path: data/test-*
language:
- en
tags:
- sentence-transformers
pretty_name: AskUbuntu
---

# Dataset Card for AskUbuntu

The AskUbuntu dataset (Lei et al., 2016) is a collection of preprocessed questions taken from AskUbuntu.com 2014 corpus dump. It also comes with 400*20 mannual annotations, marking pairs of questions as "similar" or "non-similar".
The dataset is sourced from the original [GitHub repository](https://github.com/taolei87/askubuntu).

Note that for the `train` split, the "positive" is the list of similar questions according to AskUbuntu, and "negative" is a list of randomly selected questions. For `dev` and `test`, the "positive" are human-annotated and may be empty. Unlike the original dataset, the "positive" data is *not* a subset of the "negative" data, they're mutually exclusive.

See also [sentence-transformers/askubuntu-questions](https://huggingface.co/datasets/sentence-transformers/askubuntu-questions) for all AskUbuntu questions in this dataset.

## Dataset Details

* Columns: "query", "positive", "negative"
* Column types: `str`, `list[str]`, `list[str]`
* Examples:
    ```python
    {
      "query": "system running in low graphic mode ( ubuntu without monitor )",
      "positive": [
        "getting system to boot in headless mode set-up without display problems"
      ],
      "negative": [
        "software center not progress showing mergelist error",
        "how to install google earth or draftsight for 64-bit os ?",
        "how to install a huawei ec-226 usb modem ?",
        ...
      ]
    }
    ```
* Collection strategy: Downloading the `train_random.txt`, `dev.text`, and `test.txt` files from the original source, and mapping the query IDs to the query texts.
* Deduplicated: No