Content Curation the Final Frontier: Collaborative Filtering Was Just the First Step
tl;dr Contextual machine learning will improve the quality of recommended and shared content.
[ibimage==39970==Large==http://thoughtly.co==self==ibimage_align-center]
The hungrier we are, the better everything tastes. In response to our disproportionate demand for quantity of content — our supply has been of questionable quality. Many social media updates are bland, boring, or off-topic, posted for the sole purpose of maintaining presence and feeding the content beast.
Every platform is responding to our collective hunger pains. Most are responding with a combination of behavioral and popularity-based content recommenders (collaborative filtering). They primarily utilize popularity and user interaction to drive decisions. There is certainly merit to this approach — behavioral-based product suggestions are great. Amazon for example, associates your voyeuristic enthusiasm for high-impact resistant HD cameras with the Phantom Drone 2.
Collaborative filtering fails to identify new and esoteric content with few user interactions.
In order to identify unique content at scale, we need an approach that actually reads content before recommending. Recommenders that primarily consider user interaction are indifferent to the actual substance of the text. They are over and under-inclusive, too much noise and too few gems. It's difficult to recommend content that has few user interaction data points associated with it. Thus, new content recommendations tend to be generalized noise, in contrast to the highly-precise recommendations of well established content.
The problem for the content consumer is clear — information exchange has become so congested with irrelevant noise that we’ve come to expect a homogenized blur of content in our news and social feeds. The triumph of content mediocrity.
With the deluge of content online, the challenge may be even greater for those looking to differentiate themselves as valuable contributors among their peers.
IntelliButler takes a substance first approach.
We call it substance first because it’s substance IntelliButler examines and content IntelliButler cares about.
Few content suggestors read what they recommend, but IntelliButler does.
How IntelliButler Works:
IntelliButler is a machine learning natural language processor. Exploring the inter-webs, IntelliButler evaluates written articles on a quest to offer you the smartest content. He’s not Google. He regularly examines content from several hundred sources and all content the IB community submits for textual analysis.
IntelliButler identifies potential content for distribution and consumption by comparing it to his knowledge base. If the association of an article is sufficiently strong it’s classified within a category and flagged for human review. A majority of content doesn’t meet IntelliButler’s threshold of contextual relevance and is not considered for review. Super-intelligent human beings then read through the flagged content and move the highest caliber pieces into the content inventory. We prefer quality of content to volume and think you’ll appreciate our approach too.
Substance First Benefits Everyone.
- Spend less time searching. Textual analyses based on machine learning can help you consider a more diverse spectrum of content.
Sharing reliable information allows you to carve out a reputable identity within your industry.
Submit smart content you create or discover. IntelliButler identifies your strongest content to be shared with those who will value it most.
IntelliButler recommends
- New and esoteric content with few user interactions.
By
- Actually reading content before recommending
Thoughtly is launching IntelliButler at TechCrunch Disrupt SF 2014.
[ibimage==39963==Large==http://thoughtly.co==self==ibimage_align-center]
