<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Ben Squire on Medium]]></title>
        <description><![CDATA[Stories by Ben Squire on Medium]]></description>
        <link>https://medium.com/@bensquire?source=rss-d08d1f48fc86------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*aaM0Qo36mrmt3uQn.</url>
            <title>Stories by Ben Squire on Medium</title>
            <link>https://medium.com/@bensquire?source=rss-d08d1f48fc86------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 07:07:18 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@bensquire/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[SumoDB in Neo4j: Microsoft Fabric bridging Decades of Rikishi — Part 4]]></title>
            <link>https://medium.com/neo4j/sumodb-in-neo4j-microsoft-fabric-bridging-decades-of-rikishi-part-4-b0871aa333ed?source=rss-d08d1f48fc86------2</link>
            <guid isPermaLink="false">https://medium.com/p/b0871aa333ed</guid>
            <category><![CDATA[microsoft-fabric]]></category>
            <category><![CDATA[graph-analytics]]></category>
            <category><![CDATA[japan]]></category>
            <category><![CDATA[neo4j]]></category>
            <category><![CDATA[sumo]]></category>
            <dc:creator><![CDATA[Ben Squire]]></dc:creator>
            <pubDate>Mon, 10 Aug 2026 16:38:19 GMT</pubDate>
            <atom:updated>2026-08-10T16:38:19.683Z</atom:updated>
            <content:encoded><![CDATA[<p>Nostalgia in sports is common — we often look back on previous generations with gilded lens. Ask any sumo fan whether the sport has changed and you’ll get the same answer: the big men don’t grapple like they used to. The belt battles that defined the 2000s have given way to shoving matches and quick pull-downs.</p><p>The first <a href="https://neo4j.com/blog/graph-data-science/sumodb-in-neo4j-chaining-multiple-graph-algorithms-in-snowflake-part-3/">three parts</a> of this series built a graph of professional sumo and used Neo4j to rank wrestlers by more than raw wins. Part 4 puts the claim to the test across 26 years of results, and it does the whole thing in Microsoft Fabric. The bout data lands in a Fabric Lakehouse, the graph algorithms run on Aura Graph Analytics, and the scored results flow straight back to OneLake where any Fabric tool can read them.</p><h3>What we’re building</h3><p>The pipeline has four hops, and each one hands off cleanly to the next:</p><pre>sumo-api.com  →  Fabric Lakehouse (bronze → silver → gold)  →  Aura Graph Analytics  →  OneLake</pre><p>The data comes from <a href="https://www.sumo-api.com/">sumo-api.com</a>, a free JSON API with Grand Sumo results back to 1958. We pull every bout from the top two divisions, Makuuchi and Juryo, for January 2000 through the present. That’s <strong>78,710 bouts</strong>, roughly 47,000 in the top division and 31,000 in Juryo.</p><p>Why both divisions? Juryo is the proving ground where tomorrow’s champions learn their craft. Keeping it in lets us compare the two tiers and follow wrestlers as they climb from Juryo into top-division dominance, changing their style along the way.</p><h3>From winning techniques to a style vector</h3><p>Sumo records how every bout ends with a <em>kimarite</em>, the winning technique. There are 82 of them, from the workmanlike yorikiri (force-out) to the once-a-decade izori (backward body drop). Eighty-two categories are too many to reason about, so we group them into six style families:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/697/1*MYVqRwnq48RUn9LwUgUQyg.png" /></figure><p>Every wrestler gets a <strong>style vector</strong>: the share of their wins that came from each family. A pure pusher scores high on OSHI and near zero everywhere else. A belt technician loads up on YOTSU and NAGE. This vector is the raw material the graph clusters on.</p><p>The mapping is a plain Python dictionary, so it travels with every stage of the pipeline:</p><pre>FAMILY = {<br>    &quot;oshidashi&quot;: &quot;OSHI&quot;, &quot;tsukidashi&quot;: &quot;OSHI&quot;,<br>    &quot;yorikiri&quot;: &quot;YOTSU&quot;, &quot;tsuridashi&quot;: &quot;YOTSU&quot;,<br>    &quot;uwatenage&quot;: &quot;NAGE&quot;, &quot;shitatenage&quot;: &quot;NAGE&quot;,<br>    &quot;hatakikomi&quot;: &quot;REACTIVE&quot;, &quot;hikiotoshi&quot;: &quot;REACTIVE&quot;,<br>    # ...82 techniques in total<br>}</pre><h3>Building the medallion in Fabric</h3><p>Inside the Lakehouse we follow the <a href="https://learn.microsoft.com/en-us/azure/databricks/lakehouse/medallion">standard</a> bronze, silver, gold pattern. Bronze ingests the raw scrape and types the columns. Silver does the enrichment: it attaches the style family, parses rank classes, and assigns a prestige weight so that beating a yokozuna counts for more than beating a rank-and-file maegashira.</p><pre>from pyspark.sql import functions as F<br>from itertools import chain<br><br>RANK_WEIGHT = {&quot;Y&quot;: 6.0, &quot;O&quot;: 5.0, &quot;S&quot;: 4.0, &quot;K&quot;: 3.0, &quot;M&quot;: 1.0, &quot;J&quot;: 0.5}<br>fam  = F.create_map([F.lit(x) for x in chain(*FAMILY.items())])<br>rank = F.create_map([F.lit(x) for x in chain(*RANK_WEIGHT.items())])<br>silver = (bronze<br>    .withColumn(&quot;kimarite_family&quot;,<br>                F.coalesce(fam[F.lower(&quot;kimarite&quot;)], F.lit(&quot;SPECIAL&quot;)))<br>    .withColumn(&quot;loser_rank_class&quot;, F.substring(&quot;loser_rank&quot;, 1, 1))<br>    .withColumn(&quot;loser_weight&quot;,<br>                F.coalesce(rank[F.col(&quot;loser_rank_class&quot;)], F.lit(0.5))))</pre><p>Gold rolls the bouts up into the two tables the graph needs:</p><ul><li>gold_rikishi_style, one row per wrestler with the six style shares plus career stats. These become the <strong>Rikishi nodes</strong>.</li><li>gold_defeated_edges, aggregated winner-to-loser pairs weighted by the prestige of the beaten opponent. These become the <strong>DEFEATED relationships</strong>.</li></ul><p>We filter the node table to wrestlers with at least 20 top-two-division wins, which keeps the analysis on established careers with a stable style signal. That leaves a connected core of <strong>317 wrestlers</strong> and <strong>24,066 weighted edges</strong>.</p><h3>Modeling the graph on Serverless AuraDB</h3><p>Fabric ships a native way to turn Lakehouse tables into a graph. In your workspace, choose <strong>New item → Neo4j Graph Dataset</strong>, connect an AuraDB tenant, and pick <strong>Serverless</strong> so you can drive everything from a notebook.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/737/1*WgD4Jyz48WcBcXctl-Ur_Q.png" /></figure><p>When you create your Aura instance you will be given a set of credentials. Be sure to save these somewhere safe as you will need them again later.</p><p>Select the tables you wish to import into Neo4j. In our case we will only need our gold_rikishi_style table which will generate the nodes of the graph as well as the gold_defeated_edges table which will create the edges.</p><p>Neo4j Graph Intelligence importer uses AI to help generate potential graph model based on the tables you select. You can alter the model in the graph model editor or if everything looks right you can click Transform to graph.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6Mb_zZWby4qG0o35HaP22g.png" /></figure><p>After you press Transform to graph, the import job begins. You will see the steps processing as you wait.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/292/1*u1d4R9ChZmJxljLJL4Ff2w.png" /></figure><p>When the job completes you will get notified both on screen and an email will be sent to your Fabric account email.Now you can navigate to Neo4j Graph in Fabric and query against it in Cypher through the query tab.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4kn3T3pjcG1Y56XEvdzlow.png" /></figure><h3>Running Aura Graph Analytics from a Fabric notebook</h3><p>Now the fun part. Aura Graph Analytics gives you a serverless GDS session: you spin it up for the length of the job, run algorithms on a projected graph, write the answers back, and shut it down. The power of graphs an API call away in your notebooks.</p><p>Open a Python notebook in Fabric, install the client, and start a session against your AuraDB instance:</p><pre>%pip install graphdatascience<br><br>from graphdatascience.session import (<br>    GdsSessions, AuraAPICredentials, DbmsConnectionInfo, AlgorithmCategory,<br>)<br>from datetime import timedelta<br>sessions = GdsSessions(api_credentials=AuraAPICredentials(CLIENT_ID, CLIENT_SECRET, TENANT_ID))<br>memory = sessions.estimate(<br>    node_count=1_000, relationship_count=30_000,<br>    algorithm_categories=[AlgorithmCategory.CENTRALITY,<br>                          AlgorithmCategory.COMMUNITY_DETECTION],<br>)<br>gds = sessions.get_or_create(<br>    &quot;sumo-part4&quot;, memory=memory,<br>    db_connection=DbmsConnectionInfo(NEO4J_URI, NEO4J_USERNAME, NEO4J_PASSWORD),<br>    ttl=timedelta(hours=2),<br>)</pre><p>Next, project the graph into the session. Its worth nothing a few things. We build the style vector inline from the six share properties and cast each one to a float, because a session graph rejects a list that mixes integers and floats. And we project the DEFEATED edge in reverse, from loser to winner, so that PageRank prestige flows toward the wrestlers who won the high-value bouts.</p><pre>query = &quot;&quot;&quot;<br>CALL () {<br>    MATCH (w:Rikishi)-[d:DEFEATED]-&gt;(l:Rikishi)<br>    RETURN l AS source, w AS target,<br>        { styleVector: [coalesce(toFloat(l.oshiShare),0.0), coalesce(toFloat(l.yotsuShare),0.0),<br>                        coalesce(toFloat(l.nageShare),0.0), coalesce(toFloat(l.reactiveShare),0.0),<br>                        coalesce(toFloat(l.legShare),0.0), coalesce(toFloat(l.specialShare),0.0)] }<br>            AS sourceNodeProperties,<br>        { styleVector: [coalesce(toFloat(w.oshiShare),0.0), coalesce(toFloat(w.yotsuShare),0.0),<br>                        coalesce(toFloat(w.nageShare),0.0), coalesce(toFloat(w.reactiveShare),0.0),<br>                        coalesce(toFloat(w.legShare),0.0), coalesce(toFloat(w.specialShare),0.0)] }<br>            AS targetNodeProperties,<br>        toFloat(d.weight) AS weight<br>}<br>RETURN gds.graph.project.remote(source, target, {<br>    sourceNodeLabels: labels(source), targetNodeLabels: labels(target),<br>    sourceNodeProperties: sourceNodeProperties, targetNodeProperties: targetNodeProperties,<br>    relationshipType: &#39;LOST_TO&#39;, relationshipProperties: { weight: weight }<br>})<br>&quot;&quot;&quot;<br>G, _ = gds.graph.project(graph_name=&quot;sumograph&quot;, query=query)</pre><p>With the graph in memory, run two algorithms. Rank-weighted PageRank answers “who beat the strongest opposition,” and it writes a dominanceScore back to every wrestler:</p><pre>gds.pageRank.write(<br>    G, relationshipTypes=[&quot;LOST_TO&quot;], relationshipWeightProperty=&quot;weight&quot;,<br>    dampingFactor=0.85, maxIterations=30, writeProperty=&quot;dominanceScore&quot;,<br>)</pre><p>This same algorithm can be run in the Bloom tab to show a visualization of which Rikishi are on top.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ip4en03Pjj70D00JKhDcQA.png" /></figure><p>Then <a href="https://neo4j.com/docs/graph-data-science/current/algorithms/knn/">KNN</a> builds a similarity graph from the style vectors. For each wrestler in the graph, the algorithm finds the top 15 wrestlers that have the closest similar style vector edges based on matches with other wrestlers and adds a score to those edges using Cosine Similarity. <a href="https://neo4j.com/docs/graph-data-science/current/algorithms/louvain/">Louvain Community Detection</a> then groups wrestlers who win the same way into style archetypes, this is done by maximizing the change in <a href="https://en.wikipedia.org/wiki/Modularity_(networks)">modularity</a> as separate Rikishi are added to the same community:</p><pre>gds.knn.mutate(<br>    G, nodeProperties={&quot;styleVector&quot;: &quot;COSINE&quot;}, topK=15,<br>    similarityCutoff=0.80, randomSeed=42,<br>    mutateRelationshipType=&quot;SIMILAR&quot;, mutateProperty=&quot;score&quot;,<br>)<br>gds.louvain.write(<br>    G, relationshipTypes=[&quot;SIMILAR&quot;], relationshipWeightProperty=&quot;score&quot;,<br>    writeProperty=&quot;styleArchetype&quot;,<br>)</pre><p>This can be visualized in Bloom by selecting Rule-Based Color. We will first overlay the Primary Decade a Rikishi fought in and then show the Style Archetype they were found to have based on KNN &amp; Louvain.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WnHtF_f3nRs02WFZ3X0Y8w.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*X1e2WZI49md5deW0RRuupA.png" /></figure><p>From a 10000 foot view it’s hard to tell if the overall archetypes have changed from one decade to the next. Lets write the results back to OneLake so we can do some SQL analysis to answer our starting question.</p><h3>Writing results back to OneLake</h3><p>The graph scores are useless if they stay locked in the session. Read them into a DataFrame and save them as a Delta table, and the whole of Fabric can query them:</p><pre>df = gds.run_cypher(&quot;&quot;&quot;<br>    MATCH (r:Rikishi)<br>    RETURN r.shikona AS shikona, r.primaryDecade AS decade,<br>           r.primaryDivision AS division, r.dominanceScore AS dominance,<br>           r.styleArchetype AS archetype,<br>           r.oshiShare AS oshi, r.yotsuShare AS yotsu,<br>           r.nageShare AS nage, r.reactiveShare AS reactive<br>&quot;&quot;&quot;)<br>spark.createDataFrame(df).write.format(&quot;delta&quot;).mode(&quot;overwrite&quot;) \<br>     .saveAsTable(&quot;gold_rikishi_graph_scores&quot;)</pre><p>From here the analysis is plain SQL, which means Power BI, notebooks, and the rest of your Fabric estate all read the same numbers.</p><h3>What the graph found</h3><h4>Seven style archetypes</h4><p>Louvain found seven communities, and their average style vectors give each a clear identity. The exemplars confirm them: the pushers really are the pushers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1007/1*t8tR_m2odNq4DBhVzYnl5A.png" /></figure><h4>The style mix shifts across decades</h4><p>Track each archetype’s share of its decade and the fan’s hunch turns out to be half right. The belt-and-throw technician, the wrestler who wins with a grip and a throw the way Hakuho did, falls from 23% of the 2000s to about 16% of the 2010s and 2020s. Rising to meet it is the two-way hybrid, up from 11% to nearly 18%, now the most common archetype in the sport. Wrestlers are specializing less and mixing pushing with belt work more.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/863/1*gEdQh4kHwz0RQYYOeN33ew.png" /></figure><p>The aggregate technique counts tell the same story from another angle. In the top division, OSHI’s share of wins climbs from 25% in the 2000s to 32% in the 2020s, while throwing (NAGE) slips from 14% to 12%. Pushing is up, throwing is down, and the graph shows which wrestlers drove the change.</p><h4>The shift reaches the very top</h4><p>A change in the general population is one thing. The telling test is the leaderboard, so rank the most dominant wrestlers of each decade and read off their archetypes:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*V1EqMN3RFuWoALA4u2pjQA.png" /></figure><p>Three of the five most dominant men of the 2000s win with a grip and a throw. The belt still rules through the 2010s, the Hakuho decade. By the 2020s the leaderboard belongs to pushers and two-way hybrids, and Terunofuji stands out as the last belt specialist holding the top. The style drift runs the full length of the ranking, from the champions down to the rank and file.</p><h3>The two divisions diverge</h3><p>In the 2020s, Juryo leans more on belt work than Makuuchi does, with YOTSU about five points higher, while the top division shows more reactive sumo. The lower division is the more conservative, grappling-first tier, and wrestlers adapt as they climb. Follow a wrestler across the promotion line and many shift toward belt sumo when they reach the top: Asashoryu’s win mix gained 28 points of belt-and-throw share between his Juryo and Makuuchi careers.</p><h3>Notes from the field</h3><p>A few things will trip you up if you follow this path, so here’s what to watch for:</p><ul><li><strong>The </strong><strong>%pip install lasts one session.</strong> Fabric scopes it to the current Spark session, so rerun the install cell after any timeout or restart. For a permanent fix, bake graphdatascience into a custom Environment.</li><li><strong>Relative Lakehouse paths need a default lakehouse.</strong> If Files/... resolves to a trusted-service-user folder, you forgot to pin the lakehouse as default on that notebook.</li><li><strong>The tenant id is a bare UUID.</strong> Copy it out of the console URL and drop the trailing slash, or the session call fails with a project mismatch.</li><li><strong>The import tool renames columns.</strong> OSHI_share becomes oshiShare, total_wins becomes totalWins. Run db.propertyKeys() and match your Cypher to the real names.</li><li><strong>Session graphs want uniform lists.</strong> Wrap every style-vector element in toFloat(...) so a stray integer share doesn&#39;t poison the list.</li></ul><h3>Building on your Own</h3><p>The finished pipeline reads its data from OneLake, runs graph algorithms on a session that exists only while the job runs, and writes dominance scores and style archetypes right back to OneLake as a governed table. Your Fabric stack keeps its single source of truth, and it gains the one thing tables alone can’t give you: an understanding of how every wrestler connects to every other.</p><p>And the sport? The fans were onto something. Belt sumo hasn’t vanished, but the specialist who lived on the grip is giving ground to the wrestler who can do a bit of everything and shove when he needs to.</p><p>While Sumo wrestlers and their Kimarite style may not be your focus professionally, perhaps customers and their buying behavior is. This workflow could be analogously applied to customers buying or viewing habits across decades. Looking for similar customer archetypes and analyzing how the top customers are shifting and where.</p><p>You can run this yourself for free. Start a <a href="https://neo4j.com/docs/graph-intelligence-microsoft-fabric/quick-start/">Serverless AuraDB</a> instance and an Aura Graph Analytics session on the <a href="https://console.neo4j.io/">Neo4j console</a>, point them at your own Fabric data, and see what your graph has been hiding.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b0871aa333ed" width="1" height="1" alt=""><hr><p><a href="https://medium.com/neo4j/sumodb-in-neo4j-microsoft-fabric-bridging-decades-of-rikishi-part-4-b0871aa333ed">SumoDB in Neo4j: Microsoft Fabric bridging Decades of Rikishi — Part 4</a> was originally published in <a href="https://medium.com/neo4j">Neo4j Developer Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[SumoDB in Neo4j: Chaining Multiple Graph Algorithms in Snowflake — Part 3]]></title>
            <link>https://medium.com/neo4j/sumodb-in-neo4j-chaining-multiple-graph-algorithms-in-snowflake-part-3-3c33ff130c1c?source=rss-d08d1f48fc86------2</link>
            <guid isPermaLink="false">https://medium.com/p/3c33ff130c1c</guid>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[graph-algorithms]]></category>
            <category><![CDATA[sumo]]></category>
            <category><![CDATA[neo4j]]></category>
            <category><![CDATA[snowflake]]></category>
            <dc:creator><![CDATA[Ben Squire]]></dc:creator>
            <pubDate>Tue, 19 May 2026 18:09:33 GMT</pubDate>
            <atom:updated>2026-05-19T18:10:23.697Z</atom:updated>
            <content:encoded><![CDATA[<p>Previously we <a href="https://neo4j.com/blog/graph-data-science/sumodb-in-neo4j-graph-analytics-in-snowflake-part-2/">showed</a> how its possible to run Graph Analytics directly in snowflake using our Sumo Dataset from <a href="https://sumodb.sumogames.de/Default.aspx">https://sumodb.sumogames.de/Default.aspx</a>. Our analysis highlighted how using only the structure of sumo matches we can discover the approximate rank/community of a wrestler based on the other wrestlers they faced using Louvain Community detection.</p><p>In this post, we combine Neo4j Graph Analytics with Snowflake SQL to uncover insights about sumo wrestling that neither tool could surface alone. We’ll show how PageRank reveals true competitive prestige beyond raw win counts, how Betweenness Centrality identifies the wrestlers who act as critical bridges between the division’s dominant and mid-tier competitors, and how rock-paper-scissors-cycles in dominance relationships map the non-transitive rivalries that define the current era of top-division sumo. It is prestige, structure, and chaos that produce a composite view of the competitive landscape that no single algorithm or SQL query could deliver on its own.</p><h4>The Data — Makuuchi bouts from 2021.01–2025.11</h4><p>Our data starts from gathering all bouts from Juryou and Makuuchi for the top 42 rikishi of the 2026 Haru Basho. Our original dataset was somewhat skewed as it includes one of the longest fighting Riksihi in Sumo history (Tamawashi — #1 most top division bouts of all time), because of this we will filter our analysis to Makuuchi bouts from the last 5 years. Further, we will only consider rikishi with 20 or more bouts to get better statistical significance on their fighting style and dominance record.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jfQsrkjj22EvbMg0-Cl8fw.png" /></figure><h4>Who is the most dominant Rikishi in the last 5 years?</h4><p>While raw win count is a straight-forward metric that can highlight dominance among rikishi, it fails to capture the whole story. As rikishi progress through each basho, they face a variety of different ranks, and their winning/losing record at the end of each basho determine if they will advance or be demoted (apart from Yokozuna which reach a permanent status once achieved). This means some wins are more prestigious than others, especially when the win is against the best of the best.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nFE5xgR5ChCasL2oVy4aVw.png" /></figure><p>A better metric that considers the strength of schedule is called PageRank.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/578/1*43tCcNbHUMq0W-eolnZPSQ.png" /></figure><p>PageRank in our sumo-bout network measures the quality of a wrestler’s victories, not just the count. Beating a wrestler who beat many strong opponents propagates more prestige than accumulating wins against weaker competition. This surfaces the wrestlers who truly dominated the strongest opposition in the 2021–2025 era. We can use weights to represent the prestige of each bout by the rank of the rikishi involved.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/631/1*a6tYC_Plgc4MWnfPgyhb3w.png" /></figure><h4>Defining our Input data for PageRank</h4><p>Using these rank weights we build a directed edge table with one row per bout, flowing from winner to loser. This serves as input to the weighted PageRank algorithm via <a href="https://neo4j.com/docs/snowflake-graph-analytics/current/algorithms/page-rank/">Neo4j Graph Analytics for Snowflake</a>. Weighting each edge by the loser’s rank ensures that victories over higher-ranked opponents contribute more prestige to a wrestler’s final score.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*nhr0SlsFtTje7gFP1wv9Vg.png" /></figure><p>The winner of each bout is assigned as the sourceNodeId and the loser as the targetNodeId. These column aliases are required by the PageRank procedure and must match exactly for the algorithm to recognize the edge table correctly.</p><p>One important consideration when invoking the algorithm is orientation. Our DEFEATED relationships point from winner to loser, meaning prestige would naturally flow toward losers under a standard orientation. Since we want prestige to flow toward winners, we use a reverse orientation in the algorithm call, flipping the direction so that incoming edges point toward the wrestler who won rather than the one who lost.</p><p>Finally, we create a node table with a nodeId alias, which serves as the second required input alongside the edge table when calling the PageRank procedure. As with the edge table aliases, this naming convention is enforced by the algorithm and must be consistent across both tables.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/463/1*yiCW5D9ySZFm5Z8DmZdVnQ.png" /></figure><h4>PageRank Configurable Parameters</h4><p>Two configurable parameters found in the compute section are shown, dampeningFactor and maxIterations. The damping factor models the probability that prestige continues to flow through the network versus “teleporting” to a random node. At 0.85 it means there is an 85% chance that prestige from a defeated wrestler propagates forward to their conqueror, and a 15% chance it resets to a random wrestler entirely.</p><p>In sumo terms, a wrestler gains prestige from beating strong opponents, but that prestige chain doesn’t continue indefinitely. The 15% reset prevents a single dominant chain of victories from monopolizing all the prestige in the network and ensures every wrestler retains some baseline score regardless of who beat them. 0.85 is the standard default used in most PageRank implementations including <a href="https://research.google/pubs/the-anatomy-of-a-large-scale-hypertextual-web-search-engine/">Google’s original paper</a>.</p><p>PageRank is iterative. Each pass redistributes prestige scores across the entire network based on the previous pass. Max iterations caps how many times this redistribution runs. With each iteration the scores converge closer to their final stable values, and 20 iterations is typically sufficient for a graph of this size to reach convergence. Together these parameters ensure the algorithm converges to a stable, meaningful prestige ranking rather than amplifying noise from isolated victory chains.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kjVUKJvvxP_6icqds9M5zQ.png" /></figure><p>We can now compare each rikishi’s PageRank score against their raw win count over the 2021–2025 period. Ranking by PageRank rather than wins reveals where the two diverge. A positive delta indicates a wrestler whose victories carried more prestige than their win volume alone would suggest, while a negative delta flags wrestlers who accumulated wins against comparatively weaker opposition.</p><p>At the top, Hoshoryu, Kirishima, Kotozakura, and Daieisho lead in both wins and PageRank, a strong signal of sustained dominance against quality opponents over the full five year window. Further down the table, Abi, Wakatakakage, and Takayasu tell a different story. All three have significantly fewer total bouts than the wrestlers above them, likely due to injury during the analysis window. Yet their PageRank scores remain competitive. When they competed, they beat wrestlers who mattered.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*60MA0NL6y0ltYwNw9aRxqw.png" /></figure><h4>Building a Dominance Graph: Who Owns Each Rivalry?</h4><p>Grand sumo inherently has built in rivalries by the design of the tournament structure. Rikishi regularly face each other repeatedly allowing for some wrestlers to best others in a dominant fashion over time. We can build a directed graph which analyzes which rikishi have significant win advantages over others.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/665/1*ws_D9mveXM7Z8jAD_-W5rg.png" /></figure><p>Thousands of individual bouts create a dense and noisy graph where the sheer volume of edges makes it difficult to see the underlying competitive structure. Two wrestlers who have met fifteen times, trading wins back and forth, appear as fifteen separate relationships rather than what they really represent: one wrestler with a slight but consistent edge over the other. Collapsing each matchup into a single directed edge pointing toward the net winner cuts through that noise and surfaces the true dominance relationships that define the division.</p><p>While still highly connected, relationships now show who holds the dominant advantage between each rikishi. Overlaying our PageRank algorithm we can see Hoshoryu still leads our ranking followed by Onosato, Kotonowaka, and Takakeisho, seen by the larger and darker nodes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FDU0rS3OPTjh0IgVQawjOQ.png" /></figure><p>Despite his position at the top of the PageRank rankings, Hoshoryu is far from unbeatable. Several wrestlers hold a net winning record against him, reflected by incoming DOMINATES edges in the rivalry graph. What separates him is not an absence of losses but the quality of the wrestlers he consistently beats. His outgoing DOMINATES edges point toward wrestlers who themselves dominate large portions of the division, and it is this chain of high-value victories that PageRank captures and rewards with the highest prestige score in the 2021–2025 era.</p><h4>Beyond PageRank: Finding the Division’s Structural Bridges</h4><p>PageRank tells us who wins against the best, but it doesn’t capture the structure. The wrestlers who define an era are not always the ones at the very top. Often it is the wrestlers just below them, the ones who consistently beat the mid-tier while occasionally toppling the elite. Its these rikishi who shape the competitive landscape most. If these wrestlers disappeared, the hierarchy would fracture. They are the load-bearing pillars of the division. This is what Betweenness Centrality measures.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/523/1*R4dkmtfXhwf1m4DpM8sipw.png" /></figure><p>A high Betweenness score identifies wrestlers who sit on the most paths through the dominance hierarchy. They are the wrestlers whose results connect the top of the division to the bottom. In practical terms, a high-Betweenness wrestler is one whose losses matter as much as their wins. When they lose to an elite wrestler it validates that wrestler’s dominance over the whole tier below. When they win against a lower-ranked opponent it keeps the hierarchy intact. Remove them and the competitive chain breaks.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YZAu1K0GwKsWzYLVf4QCIg.png" /></figure><p>The rikishi that are darker and larger represent those which support this heirarchy the most. They are the most common rikishi among shortest paths between each rikishi in the network.</p><h4>Rock-Paper-Scissors Sumo: Finding Non-Transitive Rivalries with Graph Analytics and defining Chaos</h4><p>An interesting aspect of the DOMINATES graph is that it highlights the rock-paper-scissors like structure built into the rikishi ecosystem.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/994/1*MJPSZW4wJbyFufeXBaHr9g.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/978/1*R9mCLUx_rtPnj18fENgGkQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*raIYmRRbApXXxbelohwRGA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1cBjnKbhNjN8LYePmnR7ng.png" /></figure><p>The above graph images show how the win margin of DOMINATES highlight that no single rikishi dominates everyone. Instead, a 3-cycle of rock-paper-scissors exists throughout at various win margin levels. When we run the Betweeness Centrality Algorithm those rikishi that are found in more of these 3-cycles will have a higher score. By using PageRank, Betweeness, and 3-cycle count we can develop a new metric, the Chaos Score.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MDnPSyjCJkeba9SlQe1HzQ.png" /></figure><p>Finding the best wrestlers in the current era of sumo turns out to be a far more complex question than it first appears. Raw win counts favor volume over quality. PageRank surfaces prestige but misses structure. Betweenness Centrality finds the structural pillars but ignores who they beat. Only by chaining these algorithms together through a combination of Neo4j Graph Analytics and Snowflake SQL does a complete picture emerge.</p><h4>Conclusion</h4><p>The wrestlers at the top of the Chaos Score are not simply the ones who won the most or beat the strongest opponents. They are the ones who did both while remaining embedded in the non-transitive rivalries that make the division genuinely competitive. That is the insight that graph analytics unlocks and that SQL alone could never surface. The dohyō has always rewarded complexity. Now we have the tools to measure it.</p><p>Next time, we will analyze how dominance and rikishi styles have changed across decades. We will utilize our Chaos score in practice to show how the types of kimarite used influences who wins and who lose.</p><p>If you enjoyed this blog, feel free to like, follow, and share it with others. You can find the code used in this blog post at my github <a href="https://github.com/SquireGraphs/bensquire-medium-blogs/tree/main/2026/sumodb-part3">here</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3c33ff130c1c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/neo4j/sumodb-in-neo4j-chaining-multiple-graph-algorithms-in-snowflake-part-3-3c33ff130c1c">SumoDB in Neo4j: Chaining Multiple Graph Algorithms in Snowflake — Part 3</a> was originally published in <a href="https://medium.com/neo4j">Neo4j Developer Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[SumoDB in Neo4j: Graph Analytics in Snowflake — Part 2]]></title>
            <link>https://medium.com/neo4j/sumodb-in-neo4j-graph-analytics-in-snowflake-part-2-0a0da5cc679a?source=rss-d08d1f48fc86------2</link>
            <guid isPermaLink="false">https://medium.com/p/0a0da5cc679a</guid>
            <category><![CDATA[snowflake]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[graph-analytics]]></category>
            <category><![CDATA[neo4j]]></category>
            <category><![CDATA[sumo]]></category>
            <dc:creator><![CDATA[Ben Squire]]></dc:creator>
            <pubDate>Tue, 31 Mar 2026 06:40:06 GMT</pubDate>
            <atom:updated>2026-05-04T19:10:03.436Z</atom:updated>
            <content:encoded><![CDATA[<h3>SumoDB in Neo4j: Graph Analytics in Snowflake — Part 2</h3><p>In our <a href="https://neo4j.com/blog/graph-data-science/sumodb-in-neo4j-graph-analytics-of-grand-sumo/">previous blog</a>, we showcased how graph databases can be used to model the world of Grand Sumo. We devised a graph model comprised of bashos, bouts, rikishi, and kimarite in Neo4j based on data from <a href="https://sumodb.sumogames.de/Default.aspx">https://sumodb.sumogames.de/Default.aspx</a>. Neo4j Bloom was used to visualize our model and we used the Neo4j Graph Data Science Algorithm Eigenvector Centrality to analyze the most important/central rikishi in all the bouts that involved the top 3 rikishi of the top division, Makuuchi.</p><p>In this post, we will dive into how to use Neo4j Graph Analytics for Snowflake. We will show how to install the Neo4j Graph Analytics App from the Snowflake Marketplace, granting appropriate permissions, and how to prepare and run Graph Algorithms directly on our native Snowflake data tables. Using Neo4j Graph Analytics for Snowflake can offer several advantages such as cloud scalability, on-demand pricing, and using the power of graphs where your data already lives.</p><h3>The Data — Makuuchi of Haru Basho 2026</h3><p>Previously we only considered the top 3 rikishi and every bout they have had in the Juryou and Makuuchi divisions. This made our graph heavily biased toward those specific rikishi. Today we will expand on our data for the graph, using the top 42 rikishi from the Makuuchi division in the Haru Basho 2026 and every bout they have had in their Juryou and Makuuchi careers. This brings us to a graph consisting of around 20k nodes and 86k relationships. Represented as a csv, we have about 25k rows and 44 columns pulled directly from the <a href="https://sumodb.sumogames.de/Default.aspx">sumodb.sumogames.de</a> website.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rvChZED-3W04kj5vx8oM8Q.png" /></figure><h3>The Platform — Snowflake</h3><p>Snowflake offers a cloud-native, fully managed data platform that is easy to use and set up. Instead of translating, transforming and loading our data into a native graph database via cypher import statements, the Neo4j-import tool, or the load CSV tool, Snowflake lets us drop our CSV in as a table and automatically infers data types without the need to map out the graph model. Our table consists of the basho date, the day of a bout, the rank and name of the West camp rikishi, the winning move (kimarite), and the rank and name of the East camp rikishi, as well as who won and lost the match.</p><h3>The Algorithm — Louvain Community Detection</h3><p>Neo4j Graph Analytics for Snowflake offers efficient and parallelized graph algorithms ranging from Dijkstra’s algorithm to graph neural networks. For this blog, we are going to dive into a famous community detection algorithm called Louvain. This algorithm works by maximizing a modularity score for each community, where the modularity quantifies the quality of an assignment of nodes to communities. This means evaluating how much more densely connected the nodes within a community are, compared to how connected they would be in a random network.</p><h3>The Setup — Getting Started</h3><p>Neo4j Graph Analytics for Snowflake can be installed from <a href="https://app.snowflake.com/marketplace/listing/GZTDZH40CN/neo4j-neo4j-graph-analytics">the Snowflake Marketplace</a>. The installation tutorial can be found <a href="https://neo4j.com/docs/snowflake-graph-analytics/current/getting-started/">here</a>. The main steps involve providing the application with appropriate privileges. This is as simple as a few calls to grant permission and usage:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jiwcVUzqVA47TQxlqkEvkg.png" /></figure><p>If the application doesn’t create the COMPUTE_POOL on installation, you can achieve this manually by doing the following:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MoDTCxIKAobtw0TWI33lOg.png" /></figure><p>Once we have the installation, permissions, and privileges set, we are ready to use Neo4j Graph Analytics on our data.</p><h3>Putting it all together</h3><p>In order to use Neo4j Graph Analytics for Snowflake, we begin by defining the node table and relationship table to be used in the Graph API. This can be seen below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/844/1*q7gW2lMNLi_EDiel4KreFA.png" /></figure><p>For our initial example, we will focus on a single Basho — Kyushu Basho 2025. We define all distinct rikishi_ids as our node table and the winning and losing rikishi id for each bout as our relationship table. Once we have our node and relationship tables created, using Louvain Community detection is done by simply calling the api with our tables in the correct places as shown below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/690/1*0aBZEjZ3X8XqTIUKS8Mkpg.png" /></figure><p>This procedure call will return the following on success:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Xdb0cjlJRcJxpxfo1MKIqA.png" /></figure><h3>The Results — Inference from Relationships</h3><p>In our example, we provided the Louvain algorithm only the Rikishi Ids and the pairs of Ids from each bout in the Kyushu Basho of 2025, yet from this data alone we are able to see some interesting results. Snowflake offers the ability to create python notebooks to visualize the graph seen <a href="https://neo4j.com/docs/snowflake-graph-analytics/current/visualization/">here</a>:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Gzg5bMlLs7nX8ZjlxzOQlA.png" /></figure><p>The three communities discovered in our example turn out to correlate to ranks of the rikishi across the basho. The upper ranks of the Makuuchi from Maegashira rank 4 up through Yokuzuna shown in purple, the rank-and-file Makuuchi ranks from Maegashira 17 through 5 can be seen in yellow, and the Juryou ranked rikishi can be seen in Orange. This makes sense as throughout the tournament most rikishi are assigned to bouts within a few ranks of their own rank.</p><h3>Conclusion</h3><p>In this post, we walked through the installation of the Neo4j Graph Analytics App from the Snowflake Marketplace, we showed how to prepare our data for the Graph API, and explained how Louvain Community Detection can be used to infer the rank of a rikishi simply by analyzing the pairings of the rikishi throughout the basho.</p><p>Next time, we will show how combining multiple graph algorithms together in Snowflake can be used to analyze the archetype of each rikishi and group rikishi by their most common kimarite to find what makes top rikishi successful.</p><p>If you enjoyed this blog, feel free to like, follow, and share it with others. You can find the code used in this blog post at my github <a href="https://github.com/SquireGraphs/bensquire-medium-blogs/tree/main/2026/sumodb-part2">here</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0a0da5cc679a" width="1" height="1" alt=""><hr><p><a href="https://medium.com/neo4j/sumodb-in-neo4j-graph-analytics-in-snowflake-part-2-0a0da5cc679a">SumoDB in Neo4j: Graph Analytics in Snowflake — Part 2</a> was originally published in <a href="https://medium.com/neo4j">Neo4j Developer Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[SumoDB in Neo4j: Graph Analytics of Grand Sumo]]></title>
            <link>https://medium.com/neo4j/sumodb-in-neo4j-graph-analytics-of-grand-sumo-25575efb93c7?source=rss-d08d1f48fc86------2</link>
            <guid isPermaLink="false">https://medium.com/p/25575efb93c7</guid>
            <category><![CDATA[osaka]]></category>
            <category><![CDATA[bloom]]></category>
            <category><![CDATA[graph-data-science]]></category>
            <category><![CDATA[sumo]]></category>
            <category><![CDATA[neo4j]]></category>
            <dc:creator><![CDATA[Ben Squire]]></dc:creator>
            <pubDate>Thu, 26 Mar 2026 22:48:36 GMT</pubDate>
            <atom:updated>2026-04-24T18:41:11.146Z</atom:updated>
            <content:encoded><![CDATA[<p>On a recent trip to Japan on a beautiful spring day in March I found myself wandering the streets of Osaka looking for something to do in the afternoon. Just my luck, the famed Grand Sumo Tournament happened to be in town for the Haru Basho — Spring tournament. As with any graph enthusiast, as I watched I wondered what kind of graph model could be built based on the unique tournament structure.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8CxvbFwxB0o_ViI1veRCMg.jpeg" /><figcaption>Day 6 of Haru Basho 2026 — Osaka, JP</figcaption></figure><h3>Modeling Grand Sumo as a Graph</h3><p>The tournament structure is quite interesting. It consists of 6 Tournaments a year in various cities across Japan. Each tournament is 15 days with 6 different divisions split into East &amp; West camps, the top two divisions being Juryo (2nd highest) and Makuuchi (top division). Juryo has 28 wrestlers or rikishi at any given time and Makuuchi consists of the best 42 rikishi in Sumo.</p><p>Each division has various rankings within them. Juryo has 14 ranks for each East and West. Makuuchi has 8–12 rikishi labelled san’yaku which are the top best 4 ranks of Yokuzuna, Ozeki, Sekiwaki and Komusubi each with a East and West distinction. The remaining 30–34 rikishi of Makuuchi are the “rank-and-file” Maegashira 1–17 also with East vs West subdivision.</p><p>A given match between two wrestlers always has a single winning move/technique called a kimarite. Uniquely, there are no weight classes in Sumo wrestling so rikishi can range anywhere from 90kg to 200kg or more in the same bout. This means weight can play a dominate factor in the outcome of a match, however speed and agility also play important roles. The rikishi with the best record in their division is the winner after 15 days, with playoff matches only used to break ties. Based on the structure of the tournament one possible way to model this system can be seen below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/946/1*P_MYKYxm81OH6BvVrpDeOg.png" /><figcaption>Graph Schema for Grand Sumo Tournaments (fig. 1)</figcaption></figure><h3>The Data of Grand Sumo</h3><p>As with most of our world these days, a vast amount of data can be found online. For the purposes of this blog all of the data was sourced from <a href="https://sumodb.sumogames.de/Default.aspx">https://sumodb.sumogames.de/Default.aspx</a>. We used python to extract HTML parsed data and transform it in to Cypher statements for use in Neo4j. For this blog, we are going to focus on the top 3 rikishi and all of their historical bouts from Juryo and Makuuchi. A sample of the data as a graph can be seen below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/870/1*cWnYzKdOUeWIvJNljUvsKg.png" /><figcaption>Sample of the data as a Graph (fig. 2)</figcaption></figure><h3>Graph Algorithms and Bloom</h3><p>The latest Neo4 Desktop App can combine the power of graph databases with graph visualizations in Bloom using Algorithms built on Graph Data Science. These can enable new insights from a 1000 ft view that otherwise might be buried in a series of tables. Below is a look at all rikishi who have ever fought Yokozuna East — Hoshoryu, Yokuzuna West — Onosato and Ozeki East — Aonishiki. We see which rikishi have defeated others with thicker lines indicating the amount of bouts shared between them during their time in Juryo and Makuuchi.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vqGQU7734Xc3vfT9T3mFpg.png" /><figcaption>Top 3 rikishi and all bouts vs other rikishi (fig. 3)</figcaption></figure><p>Suppose we want to analyze the most influential rikishis who connect the graph the most. In this case, it is those rikishi who have battled most often between Hoshoryu, Onosato and Aonishiki. Using the GDS plugin with Bloom we can overlay or project various algorithms’ scores to answer this question. One such algorithm is the Eigenvector Centrality Measure shown below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*A98_ulmjS7UJ-F8QvY19hQ.png" /><figcaption>Eigenvector Centrality Projection (fig. 4)</figcaption></figure><p>Focusing in on the top 10 scores for our Algorithm we find the 10 rikishi most influential in our graph can be seen below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NX3-gzbSHGwMbRpb0qww6w.png" /><figcaption>Top 10 most central Rikishi using Eigenvector Centrality (fig. 5)</figcaption></figure><p>One insight that jumps out is that rank doesn’t always precipitate connection among the top ranked rikishi. Despite Aonishiki being the top Ozeki -East, his centrality score is quite a bit lower than many of the other rikishi in the dataset. This can be explained by the fact that Aonishiki is one of the youngest and fastest ranking rikishi in history. In fact, had Aonishiki won the Haru Basho in Osaka this year, he likely would have been the fastest rikishi to reach Yokuzuna Rank ever at the age of 21, with his debut Basho being the July Tournament of 2023. Sadly, his race to Yokuzuna was foiled with his first professional <em>make-koshi</em>, a losing tournament record, of 7 wins and 8 losses. This means he will need to win or be runner up in two back to back tournaments to be considered for promotion to the highest rank of Yokuzuna.</p><h3>Conclusion</h3><p>In summary, we explored some of the intricacies of modeling Sumo wrestling using Graph Databases like Neo4j. We showed how tools such as Bloom and Graph Data Science Algorithms can be used to analyze the bigger picture without getting into the weeds of tables and pesky joins and counts. And we found that sumo rank doesn’t always mean you are the most central component in the sea of rikishi.</p><p>If you enjoyed this blog, feel free to like, follow, and share with others. If you would like to generate SumoDB in your own Neo4j instance, you can find the Cypher code <a href="https://github.com/SquireGraphs/bensquire-medium-blogs/tree/main/2026/sumodb-part1">here</a>. Next time, we will explore how to analyze this type of model at scale in Snowflake using the Neo4j Graph Data Science Native Snowflake App.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=25575efb93c7" width="1" height="1" alt=""><hr><p><a href="https://medium.com/neo4j/sumodb-in-neo4j-graph-analytics-of-grand-sumo-25575efb93c7">SumoDB in Neo4j: Graph Analytics of Grand Sumo</a> was originally published in <a href="https://medium.com/neo4j">Neo4j Developer Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>