Lotico New York Semantic Web Message Board › Thoughts on EVRI?
|A former member||
I came across EVRI today, and was anxious to hear the community's thoughts on it. They're are apparently a Paul Allen start-up and have a professed goal is to create the graph of semantic objects on the web. You can't search for content, you can only search for things and see how they are related or connected to other things:
Here's the description of the company from CrunchBase.
Using semantic understanding of content, evri is building the data graph of the web. The company seeks to use this to create meaningful connections without having to search.
evri users do not type queries, and do not search for content. With
evri, content finds content and presents it to the consumer. evri’s user experience and technology use semantic understanding of content to create meaningful connections among content.
For example, evri understands the difference between Will Smith the actor and Will Smith the football player, between plasma (TVs) and plasma (blood drives), and can direct consumers and advertising to the content that makes sense in context.
Here are some other blurbs about EVRI from other sites:
"Evri appears to be tackling [the semantic] problem in a different way. Other comparisons have been made to Mahalo, though Roseman said Evri analyzes content rather than using human editors."
"VentureBeat's Chris Morrison has more on Evri, saying it is not so much about search as 'Internet tourism, with an eye toward serendipitous discovery.'
New York, NY
1. Well, it appears to be a graph browser (kudos for this). Unfortunately, expansion of adjacent nodes does not seem to support user selection of predicate vocabularies (the category of expansion relationships you're interested in exploring). As a result, it is frustrating for me as a user. But this also brings up a question: Were a graph-oriented website to provide a more functional graph browser, could the general public even use it? I think the answer is yes, but it would take some creative UI engineering.
2. It also brings up another question: Given the richness of the object models for entities in the real world, (many attributes + behaviors per type), how can any website hope to do a good job of crawling the web mining semantic relationships without having a reasonably complete set of linkage names (attributes/properties) to drive the mining process? I have yet to see a technology that can discover the attributes and behaviors for a type from unstructured information and without humans in the loop.
3. How do you suppose they infer connections between entities? Based on what something inherently is? Based on pre-conceived notions of a type's attribute set (a Human would have a link to an Address and Employer)? Based on shared attribute sub-groups? Based on shared behavior (methods) sub-groups? All of these are valid ways for building graphs.
4. Does EVRI (or anyone) consider probabilistic linkages?