# CS245-2 Project 1: Graph based Recommendation — Data

来源：https://www.kaggle.com/competitions/cs245-2-project-11-graph-based-recommendation/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

<h2>File descriptions</h2>
<ul>
<li><strong>SampleSubmission.csv</strong> - The sample submission, you can check the format of expected output. </li>

Columns:

1. `index` - the row index of test set.
2. `probability` - the probability (recommandation index) of the author reading the paper. Range: [0, 1].

<li><strong>bipartite_train_ann.txt</strong> - Contains an anonymous bipartite network with two integers per line (an edge). The first id refers to an author, and the second id refers to a paper that the author cites.</li>

Columns:

1. `author` - The author id.
2. `paper` - The paper id that cited by the author.

<li><strong>bipartite_test_ann.txt</strong>The list of author-paper pairs that need to be predicted(a pair each row), and it's recommended to output your prediction in the order of pairs in this file. <strong>(Test set)</strong></li>

Columns: similar to that of <strong>bipartite_train_ann.txt</strong>.

 <li><strong>author_file_ann.txt</strong> - Contains an anonymous coauthor network with two integers (range from 0 to 6610, each represents one unique author) per row (an edge), which means two authors have cooperated.</li>

Columns:

1.2. `author` - The author id that have collaborated.

<li><strong>paper_file_ann.txt</strong> - Contains an anonymous citation network with two integers (range from 0 to 79936, each represents one unique paper) per row (an edge), referring that the paper with former id cites the paper with latter id.</li>

Columns:

1. `paper` - The paper id.
2. `paper` - The id of cited paper.

<li><strong>feature.pkl</strong> - For a better and reasonable experiment setup, we provide initial features with a dimension size of 512 for each paper, generated by USE ([Universal Sentence Encoder](https://arxiv.org/pdf/1803.11175.pdf)), and need to be read with pickle.load() .</li>

<p>In Kaggle, you just need to submit your prediction result in '.csv' format, but pay attention to other materials that need to be submitted on the canvas platform.</p>