<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>Forem: Vlad Tolm</title>
    <description>The latest articles on Forem by Vlad Tolm (@vtolm).</description>
    <link>https://forem.com/vtolm</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F2285679%2Fd30438bd-7654-4e1d-8868-cc4bc8817163.png</url>
      <title>Forem: Vlad Tolm</title>
      <link>https://forem.com/vtolm</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://forem.com/feed/vtolm"/>
    <language>en</language>
    <item>
      <title>SEO Keywords Clustering Without Magic</title>
      <dc:creator>Vlad Tolm</dc:creator>
      <pubDate>Tue, 17 Dec 2024 21:07:06 +0000</pubDate>
      <link>https://forem.com/vtolm/seo-keywords-clustering-without-magic-5885</link>
      <guid>https://forem.com/vtolm/seo-keywords-clustering-without-magic-5885</guid>
      <description>&lt;p&gt;&lt;strong&gt;I’m a founder at Potis AI, where we’re building an &lt;a href="https://www.potis.ai/" rel="noopener noreferrer"&gt;AI recruiter&lt;/a&gt;.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;SEO has always been on my back burner, but I finally decided to tackle it. Since I had zero experience beyond surface-level knowledge, I hired an agency.&lt;/p&gt;

&lt;p&gt;First thing they said? “We need to build a semantic core.” Seemed logical. They analyzed competitors and gathered a huge list of keywords.&lt;/p&gt;

&lt;p&gt;Then they told me, “Now we’ll cluster these keywords.” Still sounded straightforward.&lt;/p&gt;

&lt;p&gt;But here’s where the &lt;strong&gt;magic started&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;The agency claimed it would take &lt;strong&gt;at least two weeks&lt;/strong&gt; to cluster everything, even though there were only &lt;strong&gt;500K records&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;When I asked what algorithms or principles they used, they simply said: “We have a very big Windows server running a magical desktop program.”&lt;/p&gt;

&lt;p&gt;Fair enough, I didn’t interfere.&lt;/p&gt;

&lt;p&gt;But curiosity got the better of me. I decided to &lt;strong&gt;build my own pipeline&lt;/strong&gt; to figure out how this works.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;Filtering out noise&lt;/strong&gt;:&lt;br&gt;
I labeled 5K rows using GPT-4o-mini, marking examples of relevant and irrelevant words. Then I trained a mini-classifier. Surprise! &lt;strong&gt;90% of the data was garbage&lt;/strong&gt;.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;Building embeddings&lt;/strong&gt;:&lt;br&gt;
I used stella_en_1.5B_v5 to generate embeddings (1024 dimensions), then reduced it to 30 dimensions with LSA (Truncated SVD).&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;Clustering&lt;/strong&gt;:&lt;br&gt;
I ran multiple iterations of clustering with HDBSCAN. By the third iteration, ~70% of the data was neatly clustered. I calculated centroids and assigned the remaining elements to the nearest cluster.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;Naming clusters&lt;/strong&gt;:&lt;br&gt;
Too lazy to do it manually, I asked GPT-4o-mini to name ~2K clusters. (Sarcasm incoming: of course) most clusters revolved around one theme.&lt;/p&gt;&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;The whole process took me &lt;strong&gt;half a day and $3 on Google Colab&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;Fast forward two weeks.&lt;br&gt;
The SEO agency came back with their results: a &lt;strong&gt;dirty dataset grouped only by root words&lt;/strong&gt;. When I showed them my results, they were genuinely surprised.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Moral of the story&lt;/strong&gt;&lt;br&gt;
Not all “magic” is science. Sometimes, it’s just &lt;strong&gt;complete lack of understanding&lt;/strong&gt;.&lt;/p&gt;

</description>
      <category>seo</category>
      <category>startup</category>
    </item>
  </channel>
</rss>
