MAG for Heterogeneous Graph Learning

<p>We provide an academic graph based on a&nbsp;snapshot of the <strong>Microsoft Academic Graph</strong> from <strong>26.05.2021.</strong>&nbsp;The Microsoft Academic Graph (MAG) is a large-scale dataset containing information about scientific publication records, their citation relations, as well as authors, affiliations, journals, conferences and fields of study. We acknowledge the Microsoft Academic Graph using&nbsp;the URI&nbsp;<a href="https://aka.ms/msracad">https://aka.ms/msracad</a>. For more information regarding schema and the entities present in the original dataset please refer to: <a href="https://docs.microsoft.com/en-us/academic-services/graph/reference-data-schema">MAG schema</a>.&nbsp;</p> <p><strong>MAG for Heterogeneous Graph Learning</strong><br> We use a recent version of MAG from May 2021 and extract&nbsp;all relevant entities to build a graph that can be directly used for heterogeneous graph learning (node classification, link prediction, etc.).&nbsp;The graph contains all English papers, published after 1900, that have been cited at least 5 times per year since the time of publishing. For fairness, we set a constant citation bound of 100 for papers published before 2000. We further include two smaller subgraphs, one containing&nbsp;<em>computer science</em>&nbsp;papers and one containing&nbsp;<em>medicine</em>&nbsp;papers.&nbsp;</p> <p><em><strong>Nodes and features</strong></em><br> We define the following nodes:</p> <ul> <li><strong>paper </strong>with mag_<em>id, graph_id,&nbsp;</em><em>normalized title</em>, <em>year of publication</em>, <em>citations&nbsp;</em>and a 128-dimension <em>title embedding </em>built&nbsp;using word2vec<br> No. of papers:<em> 5,091,690</em> (all), <em>1,014,769 </em>(medicine), <em>367,576 </em>(computer science);</li> <li><strong>author&nbsp;</strong>with&nbsp;mag_<em>id, graph_id,&nbsp;normalized name,&nbsp;citations</em><br> No. of authors:&nbsp;<em>6,363,201 </em>(all),&nbsp;<em>1,797,980 </em>(medicine), <em>557,078 </em>(computer science);</li> <li><strong>field&nbsp;</strong>with&nbsp;mag_<em>id, graph_id,&nbsp;level,&nbsp;citations&nbsp;</em>denoting the hierarchical level of the field where 0 is the highest-level (e.g.&nbsp;<em>computer science</em>)<br> No. of fields:&nbsp;<em>199,457 </em>(all), <em>83,970&nbsp;</em>(medicine), <em>45,454 </em>(computer science);</li> <li><strong>affiliation&nbsp;</strong>with&nbsp;mag_<em>id, graph_id,&nbsp;citations</em><br> No. of affiliations:&nbsp;<em>19,421 </em>(all), <em>12,103 </em>(medicine), <em>10,139 </em>(computer science);&nbsp;</li> <li><strong>venue&nbsp;</strong>with&nbsp;mag_<em>id, graph_id, citations,&nbsp;type</em>&nbsp;denoting whether conference or journal<br> No. of venues:&nbsp;<em>24,608 </em>(all), <em>8,514&nbsp;</em>(medicine), <em>9,893 </em>(computer science).&nbsp;</li> </ul> <p><em><strong>Edges</strong></em><br> We define the following edges:</p> <ul> <li><strong>author </strong><em>is_affiliated_with</em><strong> affiliation</strong><br> No. of author-affiliation edges:&nbsp;<em>8,292,253 </em>(all), <em>2,265,728&nbsp;</em>(medicine), <em>665,931&nbsp;</em>(computer science);</li> <li><strong>author </strong><em>is_first/last/other </em><strong>paper</strong><br> No. of author-paper edges:&nbsp;<em>24,907,473 </em>(all), <em>5,081,752&nbsp;</em>(medicine), <em>1,269,485&nbsp;</em>(computer science);</li> <li><strong>paper </strong><em>has_citation_to</em><strong> paper</strong><br> No. of author-affiliation edges:&nbsp;<em>142,684,074 </em>(all), <em>16,808,837&nbsp;</em>(medicine), <em>4,152,804&nbsp;</em>(computer science);</li> <li><strong>paper </strong><em>conference/journal_published_at</em><strong> venue</strong><br> No. of author-affiliation edges:&nbsp;<em>5,091,690 </em>(all), <em>1,014,769&nbsp;</em>(medicine), <em>367,576&nbsp;</em>(computer science);</li> <li><strong>paper </strong><em>has_field_L0/L1/L2/L3/L4</em><strong> field</strong><br> No. of author-affiliation edges:&nbsp;<em>47,531,366 </em>(all), <em>9,403,708&nbsp;</em>(medicine), <em>3,341,395&nbsp;</em>(computer science);</li> <li><strong>field</strong><em> is_in </em><strong>field</strong><br> No. of author-affiliation edges:&nbsp;<em>339,036 </em>(all), <em>138,304&nbsp;</em>(medicine), <em>83,245&nbsp;</em>(computer science);</li> </ul> <p>We further include a reverse edge&nbsp;for each edge type defined above that is denoted with the prefix&nbsp;<em>rev_&nbsp;</em>and can be removed based on the downstream task.&nbsp;</p> <p><strong>Data structure</strong><br> The nodes and their respective features are provided as separate&nbsp;<em>.tsv</em>&nbsp;files where each feature represents a column.&nbsp;The edges are provided as a pickled python dictionary with schema:</p> <pre><code class="language-python">{target_type:     {source_type:         {edge_type:             {target_id:                 {source_id:                     {time                     }                 }             }         }     } }</code></pre> <p>We provide three compressed ZIP archives, one for each subgraph (all, medicine, computer science), however we split the file for the complete graph into 500mb chunks.&nbsp;Each archive contains the separate node features and edge dictionary.</p>