# Classifying 20 Newsgroups — Data

来源：https://www.kaggle.com/competitions/classifying-20-newsgroups/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

<p>Participants should train their models using the <strong>train.csv</strong> file, and submit the predicted labels for the messages from <strong>test.csv</strong>. The submission file should have the same format as the <strong>sample_submission.csv</strong> (i.e. 2 columns - id and predicted label for the message with respective id). Submissions should contain predicted labels for each message in the test set.</p>

<p>The dataset contains over 18000 messages with assigned topic labels, and is split into train and test parts. The training dataset contains around 11300 labelled messages and the test dataset contains 7500 messages without label (the task is to predict these labels). Below is the full list of labells and corresponding topics:</p>

<p> 0 - alt.atheism </p>
<p> 1 - comp.graphics </p>
<p> 2 - comp.os.ms-windows.misc </p>
<p> 3 - comp.sys.ibm.pc.hardware </p>
<p> 4 - comp.sys.mac.hardware </p>
<p> 5 - comp.windows.x </p>
<p> 6 - misc.forsale </p>
<p> 7 - rec.autos </p>
<p> 8 - rec.motorcycles </p>
<p> 9 - rec.sport.baseball </p>
<p> 10 - rec.sport.hockey </p>
<p> 11 - sci.crypt </p>
<p> 12 - sci.electronics </p>
<p> 13 - sci.med </p>
<p> 14 - sci.space </p>
<p> 15 - soc.religion.christian </p>
<p> 16 - talk.politics.guns </p>
<p> 17 - talk.politics.mideast </p>
<p> 18 - talk.politics.misc </p>
<p> 19 - talk.religion.misc </p>
</p>
<h2>File descriptions</h2>
<ul>
<li><strong>train.csv</strong> - the training set</li>
<li><strong>test.csv</strong> - the test set</li>
<li><strong>sample_submission.csv</strong> - a sample submission file in the correct format</li>
</ul>
<h2>Data fields</h2>
<ul>
<li><strong>id</strong> - unique id to a given message</li>
<li><strong>message</strong> - the text of a message</li>
<li><strong>topic</strong> - the label of a message's topic</li>
</ul>