I wondered how much content categorized as Style & Fashion was more specifically about smart-fit sneakers, or LGBTQ representation? How much of Health & Fitness was actually about mental health? That post-it note never left my desk.<\/p>\n\n\n\n
Because understanding these nuances is the key to creating content that stands out. <\/p>\n\n\n\n
Language processing, naturally<\/strong><\/h2>\n\n\n\nBefore we dive too deep, it\u2019s important to understand how this kind of categorization works. Parse.ly uses natural language processing<\/a> (NLP) to group content into a hierarchy of categories, sub-categories, topics, and even topic clusters, that we call \u201cstories<\/a>.\u201d Using an algorithm to parse the semantic relationships between words (aka word vectors), our co-founder Andrew Montalenti explains: <\/p>\n\n\n\n\u201cWe can tell that of the thousands of stories written about Elon Musk, some percentage are written about Elon Musk + Tesla, and some other percentage about Elon Musk + SpaceX. We can also start to learn that articles about Tesla tend to be \u201cBusiness\u201d articles, and articles about SpaceX tend to be \u201cScience\u201d articles. [\u2026] Our NLP engine learns the relationships (or non-relationships) am<\/em>ong articles in a way that models human understanding of information and content quite accurately.\u201d<\/p><\/blockquote>\n\n\n\nNaturally, a single post can be about several things, each weighted for accuracy on a scale from 0 to 1, 1 being most relevant. For now, let\u2019s only consider a category \u201crelevant\u201d if it has a score of at least 0.3.<\/p>\n\n\n\n
I analyzed just shy of 5 million urls from January through May 2019, across 360 IAB categories and discovered:<\/p>\n\n\n\n
- Only 13% of content falls squarely into one category<\/li>
- 5% is about six or more categories<\/li>
- Most content falls into 3 unique categories
<\/li><\/ul>\n\n\n\n![\"category-histogram\"](\"https:\/\/i2.wp.com\/blog.parse.ly\/wp-content\/uploads\/2019\/07\/category-histogram.png?resize=640%2C321&ssl=1\")
<\/figure><\/div>\n\n\n\nThat means a post primarily about, say, \u201cBusiness\u201d could just as likely appear in the \u201cScience\u201d category or \u201cTechnology & Computing,\u201d or both.<\/p>\n\n\n\n
So what can these cross-sections tell us about the content that really matters?<\/p>\n\n\n\n
At the intersection of interest<\/strong><\/h2>\n\n\n\nLet\u2019s assume that on any given post, the category with the highest score is the \u201cprimary\u201d category, which makes the others \u201csecondary.\u201d From there, we can study the intersections, or co-occurrences, of primary and secondary categories by interest and demand. (That is, by views and the ratio of views per post.) Let\u2019s start with the top 20 most-viewed categories of 2019. Where are the intersections of interest in our top content? And is that intersection with other top content, or something more niche? <\/p>\n\n\n\n
![\"heatmap\"](\"https:\/\/i0.wp.com\/blog.parse.ly\/wp-content\/uploads\/2019\/07\/heatmap.png?resize=950%2C813&ssl=1\")
<\/figure><\/div>\n\n\n\nClusters of color indicate high interest in related content, like the cross-section of \u201cFood & Drink\u201d and \u201cDining Out.\u201d But the peppering beyond those clusters indicates high interest to more diverse cross-sections. This is where we see content about \u201cTelevision\u201d performs best when it\u2019s crossed with \u201cMusic,\u201d \u201cNews,\u201d and \u201cTechnology & Computing.\u201d <\/p>\n\n\n\n
\u201cSports,\u201d for example, is one of the most specific categories, relating mostly to other top sports content. \u201cFood & Drink\u201d also intersects less often. <\/p>\n\n\n\n
Comparatively, traffic to \u201cNews,\u201d is highly cross-sectional, spanning across most other top categories, though \u201cLaw, Gov\u2019t & Politics\u201d and \u201cTelevision\u201d stand out.<\/p>\n\n\n\n
While much of this underscores the obvious, the extent of cross-sections (or specificity) might be surprising. In fact, some top categories don\u2019t intersect with other content at all. Of the top 20 categories, \u201cLegal Issues\u201d and \u201cWorld Soccer\u201d have the fewest intersections with other top categories. \u201cLegal Issues\u201d has comparatively little traffic, if any, coming from any other category but \u201cLaw, Gov\u2019t & Politics.\u201d <\/p>\n\n\n\n
The demand for cross-sectional content<\/strong><\/h2>\n\n\n\nBut these categories are the real gems, as they are among the most multi-faceted, cross-sectional content: <\/p>\n\n\n\n
- Technology & Computing<\/li>
- Health & Fitness<\/li>
- Business<\/li>
- Science<\/li><\/ul>\n\n\n\n
![\"tech\"](\"https:\/\/i0.wp.com\/blog.parse.ly\/wp-content\/uploads\/2019\/07\/tech.png?resize=618%2C331&ssl=1\")
<\/figure><\/div>\n\n\n\n\u201cTech & Computing\u201d intersects heavily with \u201cBusiness,\u201d \u201cScience,\u201d and \u201cLaw, Gov\u2019t & Politics.\u201d \u201cInternet Technology\u201d and \u201cGraphic Software\u201d are the first- and third-highest cross-sections by interest, with \u201cGraphics Software\u201d in high demand, perhaps illustrating the need to visually wrap our brains around the deluge of data in our day-to-day. There is also particularly high demand within Tech & Computing for gadgetry. <\/p>\n\n\n\n
Still, it seems the largest interest in content about Tech & Computing is within the industries that require it the most.<\/p>\n\n\n\n
![\"health\"](\"https:\/\/i1.wp.com\/blog.parse.ly\/wp-content\/uploads\/2019\/07\/health.png?resize=613%2C316&ssl=1\")
<\/figure><\/div>\n\n\n\nHealth & Fitness intersects highly with other top categories, \u201cPsychology\/Psychiatry\u201d and \u201cScience.\u201d \u201cFood & Drink\u201d is in particularly high demand when it comes to health content. In a way, this is much like \u201cTechnology & Computing\u201d: cross-sections with high interest may illustrate a desire to understand the powers that influence our health, where cross-sections high in demand highlight our need for the tangible control (or lack thereof) we have over it. Behind mental health and nutrition, the highest health-related category is \u201cSexuality.\u201d <\/p>\n\n\n\n
![\"business\"](\"https:\/\/i1.wp.com\/blog.parse.ly\/wp-content\/uploads\/2019\/07\/business.png?resize=596%2C307&ssl=1\")
<\/figure><\/div>\n\n\n\nNo cross-sections fascinated me more than those within \u201cBusiness.\u201d \u201cBusiness\u201d is the 5th highest category across our network, and much content in this category is also largely about \u201cLaw, Gov\u2019t & Politics\u201d or \u201cTechnology & Computing,\u201d which rank #1 and #3, respectively. The co-occurrence of these top-performers illustrates how closely linked these three domains have become; high pageviews to them all show our audience\u2019s high level of interest in these areas.<\/p>\n\n\n\n
But even this is nuanced, as \u201cMarketing\u201d and \u201cGovernment\u201d rank nearly as high. Sweeping legislative change (or lack thereof) regarding personal data and advertising continue to hold readers\u2019 attention in \u201cBusiness\u201d coverage. <\/p>\n\n\n\n
The \u201cBusiness\u201d category fascinates me because it is both sweeping and specific, both public and personal. High demand in \u201cPersonal Finance,\u201d \u201cShopping,\u201d and \u201cCredit\/Debt & Loans\u201d suggests mass appeal for how and where individual readers spend their dollars and time.<\/p>\n\n\n\n
![\"science\"](\"https:\/\/i2.wp.com\/blog.parse.ly\/wp-content\/uploads\/2019\/07\/science.png?resize=635%2C327&ssl=1\")
<\/figure><\/div>\n\n\n\nThree major cross-sections in \u201cScience\u201d include \u201cTechnology & Computing,\u201d \u201cHealth & Fitness,\u201d and \u201cLaw, Gov\u2019t & Politics.\u201d As much as these categories intersect, it would not surprise me if in 10 years, the IAB updated the categories to merge \u201cScience\u201d and \u201cTechnology & Computing\u201d into a single category. Yet each is diverse within itself. <\/p>\n\n\n\n
The top two cross-sections in science-related fields include \u201cSpace\/Astronomy\u201d and \u201cPhysics.\u201d In fact, science content about \u201cSpace\/Astronomy\u201d is also some of the highest in-demand content in this entire study. An optimist may suggest that dreams of becoming astronauts, or at least aspirations to hang with the likes of Elon Musk, is what captures our attention. But with \u201cLaw, Gov\u2019t & Politics\u201d present among science cross-sections, a pessimist may wonder if this pattern is stoked by world powers centralizing military space functions. See for yourself<\/a>.<\/p>\n\n\n\nHigh demand for \u201cWeather\u201d and \u201cGeology\u201d within science coverage remind me of how Grist covers climate change<\/a> or our study on the imbalance of disaster coverage<\/a>. <\/p>\n\n\n\nHolding the mirror up to nature<\/strong><\/h2>\n\n\n\nLanguage is messy. Humans are messy. But we\u2019re trying to make sense of it all.
Much like prescriptive linguistics, applying broad categories to \u201cstandardize\u201d swaths of diverse, rich content can limit our understanding of where we have a unique audience. Our data, indeed our attention, is as weird and messy and interesting as we are. <\/p>\n\n\n\n
If we only recognize \u201cLaw, Gov\u2019t & Politics\u201d as the most-viewed category\u2013and it is, by far, the most viewed\u2013then content creators end up in the same tired cycle of chasing big pageview numbers and flashy headlines. <\/p>\n\n\n\n
Even within the top 20 categories in our network, audiences demand specific, relatable content\u2013not the barrage of \u201cthe usual.\u201d This bodes well for anyone covering the cross-sections of these categories. <\/p>\n\n\n\n
I\u2019ll be spending a lot of time in the near future studying niche audiences. I want to know which cross-sectional categories are loosely related or hotly relevant. I want to know the main channels for generalized content, or which platforms cultivate extraordinarily unique audiences. <\/p>\n\n\n\n
by <\/strong>Kelsey Arendt<\/a><\/p>\n\n\n\n
\u201cWe can tell that of the thousands of stories written about Elon Musk, some percentage are written about Elon Musk + Tesla, and some other percentage about Elon Musk + SpaceX. We can also start to learn that articles about Tesla tend to be \u201cBusiness\u201d articles, and articles about SpaceX tend to be \u201cScience\u201d articles. [\u2026] Our NLP engine learns the relationships (or non-relationships) am<\/em>ong articles in a way that models human understanding of information and content quite accurately.\u201d<\/p><\/blockquote>\n\n\n\n
Naturally, a single post can be about several things, each weighted for accuracy on a scale from 0 to 1, 1 being most relevant. For now, let\u2019s only consider a category \u201crelevant\u201d if it has a score of at least 0.3.<\/p>\n\n\n\n
I analyzed just shy of 5 million urls from January through May 2019, across 360 IAB categories and discovered:<\/p>\n\n\n\n
- Only 13% of content falls squarely into one category<\/li>
- 5% is about six or more categories<\/li>
- Most content falls into 3 unique categories
<\/li><\/ul>\n\n\n\nThat means a post primarily about, say, \u201cBusiness\u201d could just as likely appear in the \u201cScience\u201d category or \u201cTechnology & Computing,\u201d or both.<\/p>\n\n\n\n
So what can these cross-sections tell us about the content that really matters?<\/p>\n\n\n\n
At the intersection of interest<\/strong><\/h2>\n\n\n\n
Let\u2019s assume that on any given post, the category with the highest score is the \u201cprimary\u201d category, which makes the others \u201csecondary.\u201d From there, we can study the intersections, or co-occurrences, of primary and secondary categories by interest and demand. (That is, by views and the ratio of views per post.) Let\u2019s start with the top 20 most-viewed categories of 2019. Where are the intersections of interest in our top content? And is that intersection with other top content, or something more niche? <\/p>\n\n\n\n
Clusters of color indicate high interest in related content, like the cross-section of \u201cFood & Drink\u201d and \u201cDining Out.\u201d But the peppering beyond those clusters indicates high interest to more diverse cross-sections. This is where we see content about \u201cTelevision\u201d performs best when it\u2019s crossed with \u201cMusic,\u201d \u201cNews,\u201d and \u201cTechnology & Computing.\u201d <\/p>\n\n\n\n
\u201cSports,\u201d for example, is one of the most specific categories, relating mostly to other top sports content. \u201cFood & Drink\u201d also intersects less often. <\/p>\n\n\n\n
Comparatively, traffic to \u201cNews,\u201d is highly cross-sectional, spanning across most other top categories, though \u201cLaw, Gov\u2019t & Politics\u201d and \u201cTelevision\u201d stand out.<\/p>\n\n\n\n
While much of this underscores the obvious, the extent of cross-sections (or specificity) might be surprising. In fact, some top categories don\u2019t intersect with other content at all. Of the top 20 categories, \u201cLegal Issues\u201d and \u201cWorld Soccer\u201d have the fewest intersections with other top categories. \u201cLegal Issues\u201d has comparatively little traffic, if any, coming from any other category but \u201cLaw, Gov\u2019t & Politics.\u201d <\/p>\n\n\n\n
The demand for cross-sectional content<\/strong><\/h2>\n\n\n\n
But these categories are the real gems, as they are among the most multi-faceted, cross-sectional content: <\/p>\n\n\n\n
- Technology & Computing<\/li>
- Health & Fitness<\/li>
- Business<\/li>
- Science<\/li><\/ul>\n\n\n\n
\u201cTech & Computing\u201d intersects heavily with \u201cBusiness,\u201d \u201cScience,\u201d and \u201cLaw, Gov\u2019t & Politics.\u201d \u201cInternet Technology\u201d and \u201cGraphic Software\u201d are the first- and third-highest cross-sections by interest, with \u201cGraphics Software\u201d in high demand, perhaps illustrating the need to visually wrap our brains around the deluge of data in our day-to-day. There is also particularly high demand within Tech & Computing for gadgetry. <\/p>\n\n\n\n
Still, it seems the largest interest in content about Tech & Computing is within the industries that require it the most.<\/p>\n\n\n\n
Health & Fitness intersects highly with other top categories, \u201cPsychology\/Psychiatry\u201d and \u201cScience.\u201d \u201cFood & Drink\u201d is in particularly high demand when it comes to health content. In a way, this is much like \u201cTechnology & Computing\u201d: cross-sections with high interest may illustrate a desire to understand the powers that influence our health, where cross-sections high in demand highlight our need for the tangible control (or lack thereof) we have over it. Behind mental health and nutrition, the highest health-related category is \u201cSexuality.\u201d <\/p>\n\n\n\n
No cross-sections fascinated me more than those within \u201cBusiness.\u201d \u201cBusiness\u201d is the 5th highest category across our network, and much content in this category is also largely about \u201cLaw, Gov\u2019t & Politics\u201d or \u201cTechnology & Computing,\u201d which rank #1 and #3, respectively. The co-occurrence of these top-performers illustrates how closely linked these three domains have become; high pageviews to them all show our audience\u2019s high level of interest in these areas.<\/p>\n\n\n\n
But even this is nuanced, as \u201cMarketing\u201d and \u201cGovernment\u201d rank nearly as high. Sweeping legislative change (or lack thereof) regarding personal data and advertising continue to hold readers\u2019 attention in \u201cBusiness\u201d coverage. <\/p>\n\n\n\n
The \u201cBusiness\u201d category fascinates me because it is both sweeping and specific, both public and personal. High demand in \u201cPersonal Finance,\u201d \u201cShopping,\u201d and \u201cCredit\/Debt & Loans\u201d suggests mass appeal for how and where individual readers spend their dollars and time.<\/p>\n\n\n\n
Three major cross-sections in \u201cScience\u201d include \u201cTechnology & Computing,\u201d \u201cHealth & Fitness,\u201d and \u201cLaw, Gov\u2019t & Politics.\u201d As much as these categories intersect, it would not surprise me if in 10 years, the IAB updated the categories to merge \u201cScience\u201d and \u201cTechnology & Computing\u201d into a single category. Yet each is diverse within itself. <\/p>\n\n\n\n
The top two cross-sections in science-related fields include \u201cSpace\/Astronomy\u201d and \u201cPhysics.\u201d In fact, science content about \u201cSpace\/Astronomy\u201d is also some of the highest in-demand content in this entire study. An optimist may suggest that dreams of becoming astronauts, or at least aspirations to hang with the likes of Elon Musk, is what captures our attention. But with \u201cLaw, Gov\u2019t & Politics\u201d present among science cross-sections, a pessimist may wonder if this pattern is stoked by world powers centralizing military space functions. See for yourself<\/a>.<\/p>\n\n\n\n
High demand for \u201cWeather\u201d and \u201cGeology\u201d within science coverage remind me of how Grist covers climate change<\/a> or our study on the imbalance of disaster coverage<\/a>. <\/p>\n\n\n\n
Holding the mirror up to nature<\/strong><\/h2>\n\n\n\n
Language is messy. Humans are messy. But we\u2019re trying to make sense of it all.
Much like prescriptive linguistics, applying broad categories to \u201cstandardize\u201d swaths of diverse, rich content can limit our understanding of where we have a unique audience. Our data, indeed our attention, is as weird and messy and interesting as we are. <\/p>\n\n\n\nIf we only recognize \u201cLaw, Gov\u2019t & Politics\u201d as the most-viewed category\u2013and it is, by far, the most viewed\u2013then content creators end up in the same tired cycle of chasing big pageview numbers and flashy headlines. <\/p>\n\n\n\n
Even within the top 20 categories in our network, audiences demand specific, relatable content\u2013not the barrage of \u201cthe usual.\u201d This bodes well for anyone covering the cross-sections of these categories. <\/p>\n\n\n\n
I\u2019ll be spending a lot of time in the near future studying niche audiences. I want to know which cross-sectional categories are loosely related or hotly relevant. I want to know the main channels for generalized content, or which platforms cultivate extraordinarily unique audiences. <\/p>\n\n\n\n
by <\/strong>Kelsey Arendt<\/a><\/p>\n\n\n\n