MAG for Heterogeneous Graph Learning

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