Button Button

One vision for federated search

What I'd like to see is federated deep search with friend of a friend scoring, local semantic ontologies, and hierarchical topic organization assisted by convoluted neural networks

Let's build a recipe for this mouth-watering dish.


  • Federated deep search
  • Friend of a friend scoring
  • Local semantic ontologies
  • Hierarchical topics
  • Convoluted Neural Networks


  1. Most search on the market is shallow and centralized. That's mostly fine for quick answers to some questions, but it leads to homogeneous result sets, even when those results share a trait of being deceptively unhelpful. A deep search looks for relevant results beyond the front page to get beyond SEO-oriented content. Federation is a way to share the load of operating a provider network so that the providers are accountable to the users rather than to advertisers producing the over-engineered content

  2. Friend of a friend scoring is a family of algorithms for determining how much to rely on content from a second hand source. This can be as simply as rating sources on a scale of 1-10, sharing those scores and using a geometric mean of your opinion of your friends and their opinions of their friends to determine how much to trust sources to which you are being introduced

  3. Local semantic tagging allows individual their own systems for tagging in personally significant ways. Avoiding centralized tag definitions is an important feature to defend against targeted optimizations. Semantic vocabularies can be used to define the relationships between tags from different sources

  4. Tree structures help differentiate when similar words are used in different ways, e.g. a eurogamer who's just heard of the classic board game Diplomacy will not be interested in the memoirs of Henry Kissinger, so even a person who is interested in both can have unambiguous tags for the board game and aspects of international relations. Arranging topics in hierarchies provides capabilities missing in other search patterns, but is underutilized because it is normally labor intensive

  5. Convoluted Neural Net is a machine learning algorithm that is well suited to hierarchical classification problems. Feature sets are shareable and can be locally tuned

As with all recipes, feel free to improvise and improve